Home

Robotic Sailing: Proceedings of the 4th International Robotic Sailing

Contents

1. Fig 9 Test scenarios for the evaluation including an example trajectory Vessels are shown at their start positions goals are marked with a cross The grid distance is 10m stops immediately when a solution is found By design any trajectory returned is rule compliant For each type of scenario we randomly generate 100 instances and measures the aforementioned criteria for different choices of n The planner has additionally been equipped with a time out trajectories are discarded 1f they exceed 500 seconds of simulation time or 150 control actions During node expansion 100 random actions are generated per node The results we obtain are presented in Table For every scenario we give the average over all successful planning attempts and in parentheses the respective standard deviation 6 1 Discussion Considering the success rate of planning as shown in Table l it can be seen that the performance varies between up to 90 for the easy scenarios to as low as 22 for the more difficult tasks in which precise control actions are necessary The prob abilistic nature of the PRM cannot guarantee to identify a plan in all situations However as we could not find any systematic failure the planner can be restarted eventually finding a solution Varying the amount of active nodes influences the suc cess rate as well as influencing the computational requirements This
2. Fig 3 MOOPO sailing in the Atlantic Ocean during the 2009 WRSC in Portugal www ATIBOOR ir MOOP A Miniature Sailing Robot Platform 45 Table 1 The specification of MOOPO and MOOP1 Name MOOPO and MOOPI Date of Construction Late 2008 MOOPO and early 2009 MOOP1 Sails Single Wing Sailboats Sail Actuator Servo Rudder Actuator Servo on magnetic linkage Computers Gumstix and Microchip PICI8LF4550 Wind Sensor Self made ultrasonic and rotary Compass Honeywell HMC6343 I2C solid state tilt compensated compass GPS Globalsat SiRF3 EM 408 GPS Batteries 18 2700 mAh NiMH rechargeable AA batteries Notes MOOP sailed at WRSC2009 MOOP1 at WRSC2010 almost discharged this gave an average power budget of only 2 68 Wh per day In or der to achieve this the entire control system was placed on the PIC microcontroller eliminating the power consuming Gumstix and wireless network used in MOOPO and MOOPI1 It was intended that the PIC would spend most of the time in a low power sleep mode waking only every few minutes to read the compass and set the rudder servo The GPS and wind sensor would be sampled only every few hours and only then would the target course or sail position be changed The wind sensor was a Furuno Rowind ultrasonic sensor which we had previously used in larger sailing robots where these had proven themselves to be highly accurate and reliable In order to track the robot s position a Spot satellite messenger w
3. 0422 77 Timer Initialization 02 Configure 16 bit periodic timer for wind direction 03 SysCtlPeripheralEnable SYSCTL_PERIPH_TIMERO 04 TimerConfigure TIMERO BASE TIMER_CFG_16_BIT_PAIR TIMER_CFG_A PERIODIC 05 Set the prescaler to 50 Timer Tick is 1 MHz 06 TimerPrescaleSet TIMERO BASE TIMER_A 50 07 Enable the timer 08 TimerEnable TIMERO_BASE TIMER_A 09 Load start value 10 TimerLoadSet TIMERO BASE TIMER_A OXFFFF lls GBIO Port D Initialization 12 SysCtlPeripheralEnable SYSCTL_PERIPH_GPIOD 13 Use Pin 4 as input for the wind direction 14 GPIOPinTypeGPIOInput GPIO_PORTD_BASE GPIO_PIN_4 15 Generate an interrupt on both edges of the signal 16 GPIOIntTypeSet GPIO_PORTD_BASE GPIO_PIN 4 GPIO_BOTH_EDGES IVs Tare tae ISR fOr port D pim 4 18 BSP_IntVectSet BSP_INT_ID_GPIOD BSP_GPIOD_IntHandler 19 Enable interrupt 20 BSP_IntEn BSP_INT_ID_GPIOD Fig 6 Initialization of the port reading data from the wind direction sensor AS5040 Ols BSE Tnit Imicialize BSP 02 CP lint Oy tata iene Pl 02 Osim f f Initialize uc OS i11 04 Create the Semaphores 05 LCDSem OSSemCreate 1 LCD 06 UARTSem OSSemCreate 1 UART 07 Create the start task 08 The start task creates all other tasks 09 08 err OstaskcCresatebztt RODANT 11 Start multitasking give control to uC OS II 12 2 OSStrare gt Fig 7 Initialization steps o
4. 1 Dym C L Little P Orwin E J Spjut R E Engineering Design A Project Based In troduction 3rd edn John Wiley amp Sons Inc New York 2009 2 SailBot Class Rules SailBot 5th International Robotic Sailing Competition December 23 2010 http www usna edu Users naome phmiller SailBot SailBot htm cited May 31 2011 www A TIBOOK ir A Systems Engineering Approach to the Development of an Autonomous Sailing Vessel 99 3 SailBot Sailing Instructions SailBot 5th International Robotic SailingCompetition December 23 2010 http www usna edu Users naome phmiller SailBot SailBot htm cited May 31 2011 4 World Robotic Sailing Championship Online INNOC sterreichische Gesellschaft f r innovative Computerwissenschaften 2010 http www roboticsailing org cited May 31 2011 www ATIBOOR ir www ATIBOOR ir Using ARM7 and uC OS II to Control an Autonomous Sailboat Michael Koch and Wilhelm Petersen Abstract This paper presents some aspects of an autonomous sailboat project we started at our university An interdisciplinary group of students of mechanical en gineering electrical engineering and computer science are building the boat and are developing the control hard and software Our first attempt to control the boat combines an ARM7 microcontroller and the real time operating system uC OS L Beside our first boat design we will talk about the used control hard and soft ware especia
5. Alexander Schlaefer Ole Blauro a Editors Proceedings 5 of the 4th Inter onal Robotic Sailing Conference Robotic Sailing www A TIBOOK ir www ATIBOOR ir Alexander Schlaefer and Ole Blaurock Eds Robotic Sailing Proceedings of the 4th International Robotic Sailing Conference ya Springer www ATIBOOR ir Editors Prof Dr Alexander Schlaefer Prof Dr Ole Blaurock University of Luebeck University of Applied Sciences Luebeck Institute for Robotics and Fachbereich Elektrotechnik Cognitive Systems und Informatik Ratzeburger Allee 160 M nkhofer Weg 239 23538 L beck 23562 L beck Germany E mail blaurock fh luebeck de E mail schlaefer rob uni luebeck de ISBN 978 3 642 22835 3 e ISBN 978 3 642 22836 0 DOI 10 1007 978 3 642 22836 0 Library of Congress Control Number 2011933572 2011 Springer Verlag Berlin Heidelberg This work is subject to copyright All rights are reserved whether the whole or part of the mate rial is concerned specifically the rights of translation reprinting reuse of illustrations recitation broadcasting reproduction on microfilm or in any other way and storage in data banks Dupli cation of this publication or parts thereof is permitted only under the provisions of the German Copyright Law of September 9 1965 in its current version and permission for use must always be obtained from Springer Violations are liable to prosecution under the German C
6. edu viewdoc download doi 10 1 1 122 1739 amp rep repl amp type pdf Bohnenberger J G F Beschreibung einer Maschine zur Erl uterung der Gesetze der Umdrehung der Erde um ihre Axe und der Ver nderung der Lage der letzteren T binger Bl tter f r Naturwissenschaften und Arzneikunde 3 72 83 1817 http www ion org museum files File_l1 pdf Bowditch N The American Practical Navigator Paradise Cay Publications 2010 http amazon com o ASIN 0939837544 Briere Y Iboat An autonomous robot for long term offshore operation In The 14th IEEE Mediterranean Electrotechnical Conference MELECON 2008 pp 323 329 IEEE Los Alamitos 2008 ftp www inc eng kmutt ac th pornpoj SailBoat getPDF pdf Briere Y Bastianelli F Gagneul M Cormerais P Challenge Microtransat In CETSIS 2005 Nancy France 2005 http oatao univ toulouse fr 126 1f Briere 126 pdf Bruder R Stender B Schlaefer A Model sailboats as a testbed for artificial intelli gence methods In Proceedings of the 2nd International Robotic Sailing Conference pp 37 42 2009 Burnie M ed Participant Package World Robotic Sailing Championship 2010 and International Robotic Sailing Conference 2010 Queen s University Kingston 2010 Burns R The use of artificial neural networks for the intelligent optimal control of surface ships IEEE Journal of Oceanic Engineering 20 1 65 72 1995 http 1eeexplore ieee org xpl freeabs_all jsp a
7. ss Size of scan line always 512d untested l length of scan line header always 24d untested st unknown probably status 02 valid data 18 spin up rc raw scan line counter 0 4095 a absolute angle of scan line 0 4095 0 360 scale scan radius in meters scale x 10 V 2 ux variable but unknown multiple scanlines presumable when the automatic interference rejection discards them or the unit needs to re sync mechanically Decoding the imagery is stateless as all necessary information is included in the scan line frame See Alg 1 on how the variables are used Algorithm 1 Map scanline to Cartesian pixels a polarpixels 512 Require Out put width Out put height cosa cos a 360 4096 sina sin a 360 4096 oscale Out put width 1024 omid Out put width 2 for r 0 511 do Out put pixels2D omid cosa r oscale omid sina r oscale polarpixels r end for 6 1 1 Wireless Network Transmission Issues Although the data stream uses just around MBit of net bandwidth it is not suit able for standard WiFi e g IEEE 802 11 b g because of the way it is send out Access points negotiate a separate data rate for each client depending on radio con ditions Broadcasts and multicasts however are transmitted using the lowest possible speed so that all station in range are able to receive it That speed is usually set to www A TIBOOK ir A Digital Interface for a Lowrance Broadband Radar 175 1 or 2
8. 54 Extensive research on self trimming wing sails have been carried out by Elkaim and Boyce 23 Their experiments have shown upwind progress at 20 25 deg and speeds of 60 of the Heeling is the sidewards tilt of a sailing boat usually caused by lateral wind force www A TIBOOK ir History and Recent Developments in Robotic Sailing 9 main wing sail tail wind a b Fig 3 Self trimming wing sail a side view of an arrangement with main wing sail and tail b orientation of wing sail and tail on a close hauled course true wind speed under wind speeds from 12 25 kn approximately 6 13 m s using a self trimming wing sail on a 9 1 m catamaran 3 Scientific Community and Events 3 1 Early Examples Prior to 2005 when the idea of Microtransat Challengd initiated a new era of col laborative research in robotic sailing a large number of autonomous underwater vehicles AUV have been developed 11 4 However research on autonomous sur face vehicles ASV also known as autonomous surface crafts ASC was still in its early stage and mainly focused on motorised vessels 19 28 Just a few researchers worked on fully autonomous sailing robots According to their publications these teams seemed not to be well linked to each other A few of the most noticeable early examples are described briefly here 3 1 1 Station Keeping Autonomous Mobile Platform
9. For our first tests we used a modified fuzzy control algorithm based on 3 Cur rently only the low level functions for capturing the sensor data and controlling the actuators are embedded software modules running on the LM3S6965 whereas all high level software components are implemented on a laptop computer connected via Bluetooth The laptop gets the sensor data and sends commands back to the boat Depending of the distance the laptop 1s also used as remote control In the final set up all software components are running on the LM3S6965 the laptop is only used to show the status of the boat 5 Conclusion We have shown our approach building an autonomous sailboat In June the hull of the boat is laminated The low level software components are ready the high level software components are work in progress The work on both parts the boat and the controlling software gets on but we will still have to do many tests and improvements Our goal is the participation of our FHsailbot at the WRSC2011 in L beck www ATIBOOR ir 112 M Koch and W Petersen References 1 Ammann N Hartmann F Jauer P Bruder R Schlaefer A Design of a robotic sail boat for WRSC SailBot In IRSC 2010 Canada 2010 2 Labrosse J J MicroC OS II The Real Time Kernel CMP Books San Francisco 2002 3 Stelzer R Pr ll T John R Fuzzy logic control system for autonomous sailboats In IEEE International Conference on Fuzzy Syste
10. The single piece deck required more screws to be undone in order to access the inside of the boat caused the sail assembly to be lifted off with the deck but made access to the electronics far easier than with MOOPO The control system software for MOOPO and MOOP 1 was split into two portions A low level layer was run on the PIC18LF4550 microcontroller and interfaced to the servos compass wind sensor and GPS It received commands via a 4800 bps software serial port from the Gumstix Single Board Computer which was running a Linux based operating system Through this interface the Gumstix could send target positions for the servos or read data from the compass GPS or wind sensor The main control loop would read the compass and set the rudder at a frequency of approximately 8 10 Hz the wind sensor would be read and the sail set at a frequency of 0 33 Hz The high frequency response rate for the rudder was required to keep the boat on course and to counter a tendency to turn towards the wind during gusts Although it would have been possible to implement the control system entirely on the PIC the Gumstix provided the advantages of easy remote access and logging of data to a filesystem A small debugging LCD screen was also connected to the PIC and would report which command had been sent and its response The Gumstix was also able to send custom messages to this to show when it had finished booting or shutting down On MOOPO a Compact Flash 802
11. Without this approach much of the time spent deploying BeagleB would need to be spent debugging low level code which differs from the Tracksail version rather than actually running scientific experiments www A TIBOOK ir Simulating Sailing Robots 119 Sail PIC Rudder Current Voltage proportional to current Actuator controls Actuator Teancaucers actuator positions i UDP Packet Containing power PIC18F4550 Target Positions consumption from sent over transducers solar Parallel bus panel Serial Serial UDP Packet Containing actuator positions Voltage proportiona Sensor Compass a current USB Serial USB Serial USB Serial Converter Converter Converter Laptop PC Tracksail a PIC18F4550 NMEA and 8 bit noise re Solar representa Containing wind 4 simulated direction boat heading Parallel port solar panel and position J current Fig 3 A figure showing the dataflows used by the Hardware in the Loop Simulation Com pass GPS and Wind data can either be generated by real sensors or from the laptop running Tracksail The three selector switches determine which inputs are used by the Gumstix The Gumstix calculates actuator positions depending upon the wind GPS and compass data These positions are sent to the PIC which positio
12. and B x y the position of the waypoint and the orientation of the robot given by compass measurements for example Consider the following vectors ug cos sin 0 va Denote by u 1 1 the command associated to the rudder control u 0 means a straight rudder 1 rotated to the max position counterclockwise and 1 rotated to the max position clockwise The command u will be u K x det uyp ue 10 This strategy works but due to the drift the robot will be facing the wind when approaching the waypoint B and eventually slow down see trajectory a on Figure 9 We propose a hybrid approach which consists on following two different rules and not one The first rule is to follow the direction AB This rule is not sufficient to reach waypoint B because of the drift see trajectory c on Figure 9 The second rule is to minimize the distance e between the boat and the line AB As such the www ATIBOOR ir Sailing without Wind Sensor and Other Hardware and Software Innovations 35 Fig 9 Simulation results of three different algorithms boat will have a tendency to sail parallely to the line AB Also the boat will not loose speed arriving at the waypoint B see trajectory c on Figure 9 Consider the vector T 11 el D gt es UAB The distance e between the robot and the line AB is expressed as e det uaR AM 12 The basic hybrid command u can thus be defined by
13. connected to the deck This allowed all electronic components to be lifted out of the www A TIBOOK ir 48 C Sauz and M Neal Fig 7 A photograph of MOOP3 during radio con trolled tests Fig 8 A photograph of MOOPn both in the laboratory and sailing in Llyn Yr Orefa during its acceptance testing Table 4 The specification of MOOPn Name MOOPn Date of Construction March 2010 Sails Twin Wing Sail Sail Actuator Modified 360 degree servos Rudder Actuator Servo on magnetic linkage Computers PIC18LF4550 and Gumstix Wind Sensor Rotary wind sensor Batteries 18 2700 mAh NiMH rechargeable size AA batteries Notes Sold to Nottingham University www A TIBOOK ir MOOP A Miniature Sailing Robot Platform 49 boat as one unit which eased diagnostics and repairs A photograph of MOOPn is shown in Figure 8land full specifications are shown in Table 4 3 Experiments and Results 3 1 Lake Tests in Aberystwyth 3 1 1 Power Management Experiments A series of experiments were carried out using MOOPO and MOOP 1 at a small lake called Llyn Yr Oerfa 52 4 degrees North 3 87 degrees West approximately 20km East of Aberystwyth Due to the local topography these were primarily carried out on days with easterly or westerly winds as other wind conditions would lead to rapidly varying wind directions across the lake The primary aim of these experi ments was to develop power management algorithms based upon an abstraction of the mamma
14. current 500 mA at 5 V Sail winch electronic 10mA at 5V 150 Sail winch motor max current 3 5 A at 7 2 V Failsafe system incl remote control receiver 30 mA at 5V 50 The boat needs a minimum total current of about 360 mA and a maximum total current during the sail adjustment of about 4 A To supply the necessary current the boat is equipped with two accumulators one for the control system and the sensors regulated to 5 V and another directly connected to the sail winch The cost for the whole control system including the accumulators sensors and actuators will be less than 1000 3 5 Communication Today there are two ways to communicate with the boat During the development of the software we use the USB interface to program and to debug the system and to exchange data using a virtual COM port on the host system To get data from the boat and send data to the boat during sailing we use a Bluetooth connec tion The Bluetooth module connects to UART2 and transmits the serial data The BlueSMiRF RP SMA Sparkfun USA connects to a standard 2 4 GHz antenna On the host side we use aPCMCIA Bluetooth module with the same antenna With this combination we reach distances of about 100 m For a failsafe operation the PWM signals of the controller may be overwritten using a standard remote control RC With a separate channel of the RC we can switch between the PWM signals of the RC and the controller signals Other communication m
15. shows the system architecture of the control system Beside the con troller board with display and SD card the sensors GPS compass wind direction are providing the necessary environmental data rudder servo and sail winch are di rected by the controller The failsafe module is able to overwrite the controller sig nals of rudder and winch using a remote control to avoid problems e g collisions with other boats if the controller does not work as intended 3 1 Controller All ARM7 microcontrollers of the Stellaris Cortex M3 family from TI use the same core running at 50 MHz with different peripherals Due to the three UART we de cided to use the LM3S6965 controller If a separation of some peripherals from the main controller is necessary e g due to limited wire length of some sensors a con troller with CAN bus may be used to solve the problem Figure B shows a possible solution LM3S8962 combined with the smaller LM3S2110 The sensor on the IC port may be placed as far away as possible from metal components of a boat The ports of the LM3S2110 may be controlled over the CAN bus from the LM3S8962 www A TIBOOK ir 104 M Koch and W Petersen Hardware Bluetooth FHSailbot Host Interface 5 11 HANU m UART2 LEO so card UARTO N LIART ARTI Ce os re ee Interface Controller DEBUG Accu 7 2V 7 4V r5 TI Cortex M3 ARM LM3S6965 Compass HMC6343 lac Winddirection Rudder Servo AS5040 PWM Win
16. the solar panels and it contains all components as we can see in Fig 4 e In this area the movable portion of the actuators is placed in a watertight com partment almost like the one for Breizh Spirit 1 in which water can enter only by the outputs of the sheets e The servo motors and any part of electronics are placed in a fully watertight compartment as we can see on Fig 2 And Fig 4 For this we chose to use IP 67 certified boxes which are fully waterproof and suitable for use in very hostile marine environments The Brest Morgat test also demonstrated that all that exceed outside of the hull may be torn by waves or collisions Indeed on the Breizh Spirit 1 we chose to put the actuators and the sheets outside Both have been uprooted and lost during the Brest Morgat test Therefore on Breizh Spirit 3 we chose to protect these elements As a matter of fact the box containing the actuators is placed under the bridge just be hind the stern to be fully protected from waves and water infiltrations The circuit of sheet of Breizh Spirit 1 which was around the boat had shown good reliability particularly because the sheets didn t bring water inside the hull We have therefore chosen to design a similar system Moreover the entire circuit is placed in machined channels in the hull that allow us to protect the whole circuit with solar panels Fig 4 and Fig 5 The rudders are controlled in the same manner which limits water infiltrati
17. used similar formulas although the manual method included suggestions included in Coastal Pilots as well as cruising guides The process began with plotting courses on a gnomonic chart A gnomonic chart contains a projection of the earth which does not distort the continents or oceans This allows straight rhumb lines to be plotted that cross all meridians of longitude at the same angle giving us the shortest courses From this chart waypoints were transposed onto the distorted but more commonly used Mercator projection chart On the Mercator chart the rhumb lines were converted into great circle routes www A TIBOOK ir 186 P Gibbons Neff and P Miller sciany PILOT CHART OF THE NORTH ATLANTIC OCEAN REAR REA E AE a E ie If ae 2 4 a A y 7 sd AAY a 1 ew _ A y J 4 f f h il A r 4 x ie 26 ale Fi ME 7 mI LH F o Kp AW f a A 4 r 2 Me ae Jp je en PAS INS r 2 2 Fig 1 NOAA Pilot Chart NOAA 4 Long Term Planning A NOAA North Atlantic Pilot Chart Figure 1 2 which is a Mercator chart was used for plotting the great circle routes and the long term planning There is a pilot chart for each month which presents the average of different factors necessary for preparing an ocean crossing Each cell of size five degrees of latitude wide by five degrees of longitude high have a wind rose that represents the average wind force and percentage of direction fr
18. 60 100 50 150 100 50 50 apparent wind compass heading GPS heading A Schlaefer et al 10 15 20 time s 10 15 20 time s gyroscope deg s accelerometer y GPS speed cm s 10 15 20 time s Finally Figure l0lillustrates the tacking performance of our prototype The figure presents a situation during the aforementioned tests where the target apparent wind angle was set to 60 Starting on a stable course with virtually no rudder movements and the main and jib at angles of approximately 55 and 40 respectively the tack was initiated at about 9s The top plot clearly shows how the rudder turns the boat while the main sail moves along with the apparent wind Concurrently the jib angle www A TIBOOK ir A New Class for Robotic Sailing The Robotic Racing Micro Magic 83 first increases and then maintains an offset until the boat heads straight into the wind This is shown in the center plot which also demonstrates that the compass readings change immediately and the GPS heading is subject to a substantial delay The delay also affects the speed over ground given in the bottom plot Apparently the boat came to a complete stop during the tack which took approximately 4s from stable course to stable course 6 Conclusion We have proposed a novel one design class for robotic sailing Key features of the rrMM are the small and lightweight d
19. Approach To implement a collision avoidance algorithm only the information from the boats in a certain radius around the own position is needed This is a perfect scenario to use a broadcast approach Moreover the boats would be completely autonomous 1 e the system works without a server managing the communication process However for boats with limited resources it is a great effort to send receive and process all the broadcast messages Furthermore it is difficult to summarize the state of the race or to set up a judge or race committee www A TIBOOK ir Collision Avoidance Using a World Server 161 Table 1 Data rate and range of WLAN IEEE 802 11n and enhanced Bluetooth IEEE 802 15 1 Free2Move Sweden as communication link WLAN Bluetooth Data Rate 600 Mbit s 3 Mbit s Range 250m 500m 3 1 2 Centralized Approach In contrast to broadcasting this approach is based on collecting all data by a central server which provides it to all boats This would be particularly useful if the boats maintained a communication link to an onshore computer anyway Additionally this setup enables a global awareness of the other boats Because all data from the boats is forwarded to the world server the clients obtain a complete state of the world e g allowing to consider different race strategies Moreover it is simple to check wether all boats actually send their data Clearly this approach would be limited to all boats staying in the v
20. Bars F L Roncin K Aizier B Jaulin L Autonomous robotic boat of ensieta In IRSC 2009 Matosinhos Portugal 2009 Stelzer R Proll T John R Fuzzy logic control system for autonomous sailboats In IEEE International Conference on Fuzzy Systems 2007 www A TIBOOK ir MOOP A Miniature Sailing Robot Platform Colin Sauz and Mark Neal Abstract MOOPs Miniature Ocean Observation Platforms are small low cost lightweight sailing robots intended to form a simple and flexible platform for de veloping robotic sailing concepts and for entering the Microtransat Challenge They are only 72 cm long weigh 4 kg and can be easily transported and deployed A variety of different configurations have been tested with the intention of testing new wind sensor designs performing experiments in autonomous power management rudderless control and discovering wind direction without a wind sensor Tests have been performed in lakes and at sea During this process several limitations have been identified which cause difficulty sailing upwind difficulty sailing downwind in single wing sail models and low hull speeds which make cause problems when fighting strong currents or tides Future work will focus on more ocean going sailing and methods to reduce overall power consumption 1 Introduction This paper details the design construction and testing of sailing robots known as MOOPs Miniature Ocean Observation Platforms These robots onl
21. External PWM Servo Magnetic Status Backup External SPI 6 10V Rate Gyros Supplies Interrupts ControlPorts FieldSensor Indicator Battery Port Wireless Xbee 3 Axis 8 Channels I2C GPIO 8 Bit Parallel 3 Logic Quadrature External Lassen iQ Transceiver MMA7260Q 12 bit 0 5V GPIO LevelSerial EncoderInput 5V and 3 3V 12 Channel Accelerometer ADCInputs Ports Port Output Power GPS Receiver Under Yaw Gyro Fig 1 The NavBoard3 layout The board main face is shown on the left with the view of the underside on the right Green items indicate a component that cannot be seen in the given view occluded by other components 3 Conclusions Robotic sailing is a complex interdisciplinary task that requires significant effort in real design While competition driven it is possible to achieve high quality pe dagogical results using a systematic design methodology and appropriately scoped deliverables and metrics In this paper we have discussed an appropriate set of de sign tools and tactics for system design We have given details of our interdisci plinary interaction model which has been developed over four years of running this project The best recommendation that we can provide is to generate each ves sel as a two stage process with one year of effort on naval architecture followed by a year of system design While this is challenging the rewards are real and substantial and actually improve the quality of the design process References
22. Frank 141 Fr hwirth Thom 195 Gallou Yvon 55 Gal Oren 127 Gibbons Neff Peter 183 Hartmann Florian 157 Heinig Maximilian 71 Jafarmadar Kar m 3 Jauer Philipp 157 Jaulin Luc 27 Keef Cody 87 Koch Michael 101 Kreutzmann Arne 141 Kr ger Julia 157 Langbein Johannes 195 Leloup Richard 55 Le Pivert Frederic 55 Meyer Tobias 157 Miller Paul 183 Neal Mark 39 113 Nicola Jeremy 27 Petersen Wilhelm 101 Roncin Kostia 55 Sauze Colin 39 113 Schlaefer Alexander 71 157 Sliwka Jan 27 Stelzer Roland 3 169 195 Taschner Nicholas 87 Thomas Sebastien 55 Vienney Laurent 55 Wolter Diedrich 141 www ATIBOOR ir
23. Intelligence AI abilities using low cost sensors Our algorithm is not based on obstacles characterization Algorithm performances were tested in various scenarios with real time USV s video streams indicating that the algorithm can be used for real time applications with high success rate and fast time computation 1 Introduction One of the most difficult challenges for USV navigation is to recognize and identify obstacles around the vehicle without human intervention This task is known as Automatic Target Detection ATD An efficient ATD system should achieve high detection percentage for targets while maintaining a minimal false alarm rate This means that it must preserve an optimal balance between a high detection rate and a low error probability Although ATD algorithms are very sensitive and unstable regarding clutter ele ments 1 e elements that are not targets but still part of the scenes with similar char acteristics as the targets Dealing with clutter in ATD algorithms was extensively studied l Oren Gal Technion Israel Institute of Technology e mail orengal tx technion ac il www ATIBOOR ir 128 O Gal One of the ATD algorithms methods is based on the target temperature The contrast of the target were based on environment gradient of the target and the environment s contrast to recognize targets These methods suffer from false alarms due to targets and environment similarity Several methods were developed
24. Roadmap Planner In our approach action selection is performed by a probabilistic or randomized roadmap planner PRM 8 This type of planners is particularly helpful for mo tion planning when no inverse kinematic model is given only a forward kinematic or simulation is needed Another feature of interest is its ability to incorporate fur ther constraints such as scoring solutions by the intermediate locations visited or www A TIBOOK ir Rule Compliant Navigation with Qualitative Spatial Reasoning 149 efficiently re computing paths in dynamic environments 2 1 In a nutshell a PRM builds a graph of the search space similar to classical AI search techniques Nodes in the graph represent states of the search space they are linked by edges that are labeled by the action that allow an agent to get from one state to the next The ob jective is to determine a path from the start node to a goal node During planning a node is randomly selected and expanded by performing a fixed number of random expansions i e random actions are performed A heuristic scoring function h is em ployed to rate the expansion probability of a node and to facilitate goal directedness We represent the dynamic state of the vessel as nodes which are then linked by the rudder and sail actions performed Also we record the complete trajectory from the start position to the respective node as well as the total plan duration measured in simulation time Every node re
25. September 7 9 University of Nottingham 2009 7 Bruder R Schlaefer A Stender B Model sailboats as a testbed for artificial intelli gence methods In Proceedings of the 2nd International Robotic Sailing Conference pp 37 42 2009 8 Elkaim G Boyce C An energy scavenging autonomous surface vehicle for littoral surveillance In Proceedings of ION Global Navigation Satellite Systems Conference 2008 9 Neal M A hardware proof of concept of a sailing robot for ocean observation IEEE Transactions on Oceanic Engineering 31 2 462 469 2006 10 Neal M Sauze C Thomas B Alves J C Technologies for autonomous sailing Wings and wind sensors In Proceedings of the 2nd International Robotic Sailing Con ference Matosinhos Portugal July 6 12 pp 23 30 2009 11 Sauze C Sailing robot route planner Microtransat Challenge Sourceforge Project 2009 http microtransat svn sourceforge net viewvc ansat route_planner accessed June 10 2011 12 Sauz C A neuro endocrine inspired approach to power management in sailing robots Ph D thesis University of Wales Aberystwyth 2010 13 Sliwka J Reilhac P Leloup R Crepier P Malet H D Sittaramane P Bars F L Roncin K Aizier B Jaulin L Autonomous robotic boat of ensieta In Proceedings of the 2nd International Robotic Sailing Conference Matosinhos Portugal July 6 12 pp 1 8 2009 14 Xiao K Sliwka J L Jaulin K R A wind i
26. Subscribes Multicast Address Radar WinCE Perry babe PL i LETTI ELF Bus Puna Pena anee saien a PFLLL niset TE TT 7 Ma E ry a a a a G r e a 1 a 239 255 255 250 1900 UPNP Fig 3 Reconstructed multicast streams of a typical operation session Port numbers are de rived from actual sent data because IGMP multicast subscriptions do not carry port informa tion 6 1 Radar Image Data Stream This stream provides radar scanline data as soon as the radar is turned on by setting the appropriate control registers 0x01 and 0x02 One UDP frame containing 32 scanlines is 17160 bytes long which get fragmented by Ethernet into 1500 byte pieces default Ethernet MTU For payload data structure see table Multi byte fields are transmitted in little endian The resolution of a full 360 scan see figure 4 is 512 pixels by approx 2048 scanlines as field a is incremented by two The radar sometimes skips single or C 2048 r 512px a ib Fig 4 Radar imagery in Cartesian form a and in its original polar form b www ATIBOOR ir 174 A Dabrowski S Busch and R Stelzer Table 1 Image data packet data structure TE 00 m Frame Header ei u 73 OB m Scanline Header 5 7 55 OC 00 01 a scale 44159 13 40 CB lt 32 ul u2 Scanline 512 polar pixels in 8 bit grayscale ns Number of scan lines always 32d untested
27. Usage and Future Work Test data suggests that a strong preprocessing is necessary as radar artefacts inter ference and reflections can disturb the imagery As discussed image data from the radar is presented in polar pixels relative to the vessel This makes it easy use it in a vessel steering algorithms which use the same coordinate system The reactive approach by Sauze and Neal 9 can be applied directly on the data as each radar scan line can be used instead of a raycast For the radar to be used in the sweep line algorithm by Stelzer et al radar data first has to be converted into objects representing the outline of identified 7 The native packet capture file format of libpcap and WinPcap which Wireshark and other tools are built upon www A TIBOOK ir 180 A Dabrowski S Busch and R Stelzer obstacles i e polygons in a Cartesian like map Together with the ship s sailing wind capabilities transponder obstacles e g AIS and other map data this is used to generate a polar diagram of preferable directions to go 10 Fig 7 In that diagram every obstacle creates a dent as it makes that direction less attractive It is the base for decisions of the Short Course Routing layer To facilitate this the mapping module is currently extended to accept and deliver data in different formats and coordinate systems transparently hiding the conversion process Formats are Cartesian coordinates relative to map origin and polar coord
28. Voyage 193 AOS MNONNONwwa IN eee IN H i AN Mu Do i im Ir y gt D roe z Mh ur x a na N WY im o GEW Be ae H a a 2000 jo senza AR E IA aim A F Fig 7 Northern Route optimization for SOA using 11APR2011 GRIB file Each red line indicates an evaluated course The high density of courses near the rhumb line result from the optimization routine ati ha mu Fig 8 Southern Route optimization for SOA using 11APR2011 GRIB file Each red line indicates an evaluated course The high density of courses near the rhumb line result from the optimization routine www ATIBOOR ir 194 P Gibbons Neff and P Miller GRIB files based on short term weather models from the Global Forecast System GFS Multiple courses are analyzed using the vessel s VPP results and the course resulting in the shortest time is selected The entire course can then be programmed in to the autopilot before departure Figure 7 which was generated in Expedition for Spirit of Annapolis shows an example from the Expedition software based on a two week forecast for the Northern Route Possibly due to the slow speed of SOA the proposed route does not differ more than 75 miles from the Great Circle route Figure 8 is a similar plot for the Southern Route Recognizing the long duration of the voyage compared to the accuracy of two week models the routing can be updated onboard from satellite dow
29. Work Future work needs to focus on further improvements to the simulation model It may be possible to bring in some of the models proposed by Roncin or Riddler Vermeulen and Keuning or to make use of more realistic commercial systems Tracksail s simulation is based around a traditional fabric sail rather than an actua tor controlled wing sail This causes a number of behavioural differences compared with BeagelB for example the sail will automatically jibe as the stern turns through the wind while the wing sail must be actively turned to achieve this The simulator could be adjusted to match the behaviour of a wing sail to give more accurate perfor mance Simulations of tides and currents could be introduced Wind data could be based upon either real time feeds from an outdoor weather station generated upon average data from internet sites or replayed from previously recorded data Another possibility is to produce large lookup tables based upon actual performance data of a real sailing robot operating at sea Hardware in the Loop simulations could be made more realistic by adding weights to the actuators to simulate the presence of loads against them It might even be possible to run Hardware in the Loop Simulations on a full sailing robot while it is moored This would result in the robot experienc ing the friction of moving actuators and moving them against the wind and water Further work needs to be undertaken to complete the production o
30. a com mitment to the task and a solid plan Students can be taught the importance of planning as this project progresses especially if the deliverables include a full set of test results as discussed previously When a week by week plan of action and milestones POA amp M is completed as part of the preparatory course students do not appreciate how much effort each of their tasks will take We require our stu dents to update the POA amp M every week and provide discussion on their progress This allows the advisor to show the students how to generate optimality in the task order and how to better estimate actual required time This is a skill that we find cannot be taught in a lecture but which students rapidly appreciate when the reali ties of a complex task loom large and their grade is in the balance To assist us at USNA there are actually formal marking periods at the 6 and 12 week points in the semester offering good targets for intermediate deliverables and maintaining a steady pressure on the students across the time of the project 2 System Architecture of USNA SailBots As mentioned all of the vessels used by the USNA team for SailBot and WRSC are designed and constructed by students in the Naval Architecture department Systems Engineering students primarily focus on selection and design of the pow er sensing control and communication subsystems and to some lesser extent the actuation Over the last four years the teams have de
31. a length of more than 2 m a IBoat ISAE b FASt University of Porto c Pinta University of Aberystwyth d Beagle B University of Aberystwyth e ASV Roboat INNOC f Avalon ETH Zurich severe heeling Including batteries the overall weight of the boat is about 300 kg The sail area of mainsail and foresail together is 4 5 m7 It is equipped with solar panels providing up to 285 W of power during conditions of full sun and a direct methanol fuel cell delivering 65 W as a backup energy source The ASV Roboat fea tures a three stage communication system combining WLAN UMTS GPRS and an IRIDIUM satellite communication system allowing continuous real time access from shore 48 Control software runs on a Linux based on board computer system using incoming data from various sensors GPS compass anemometer etc on an NMEA2000 bus ASV Roboat won the Microtransat 2007 as well as the WRSC in 2008 2009 and 2010 www A TIBOOK ir 16 R Stelzer and K Jafarmadar 3 3 2 Ecole Nationale Sup rieure de Techniques Avanc es ENSTA de Bretagne A team of ENSTA Bretagne participated with their boat Breizh Spirit Figure amp a 9 in WRSC 2009 Their design uses a custom built hull based on the IMOCA class design with a length of 1 3 m and two traditional sails The control system is imple mented on a PIC18F2550 microcontroller 44 3 3 3 Institut Sup rieur de l a amp ronautique et de l espace ISAE IBoat from ISAE France Figur
32. above a user configurable limit are omitted as well Figure illustrates a portion of such a routing graph Each node in the graph is annotated with wind vectors shown as bold arrows in the figure In most cases we have multiple forecasts at hand therefore each node is annotated with one wind vector per forecast As the location of a node usually does not coincide with a grid point in the GRIB file bi linear interpolation is used to cal culate the wind vectors for each node The directed edges of the graph are annotated with the travel time calculated as shown in Section 2 1 2 Again when multiple forecasts are present each edge is annotated with one travel time per forecast The described discretization of the search space means that the optimal route in the grid model is calculated which is only an approximation to the true optimal route Thus we made the quality of the approximation configurable to the user in that the distance of the nodes in the routing graph can be arbitrarily chosen 2 2 Calculating the Optimal Route We chose the A algorithm as basis for our long term routing as it allows the use of a heuristics for performance gain 5 Much like Dijkstra s algorithm the A algorithm assigns each node x a cost value c x which is the time required to travel www A TIBOOK ir A Rule Based Approach to Long Term Routing for Autonomous Sailboats 199 to this node from the starting point The algorithm retains an open list
33. and weight Budget Because the vessels we use are optimized for sail racing the budget aspect of the systems design goes well beyond the understood monetary level to include weight power and volume Because total battery capacity weight and volume are often related tradeoffs must be made to select appropriate batteries These decisions can impact available sensing and actuation technologies Design to a four part budget is a great exercise in optimization and planning Communication Because this is a multi disciplinary team comprising systems engineers as well as naval architects communication is especially challenging The two disparate engineering disciplines do not necessarily speak the same lan guage nor mutually understand the requirements of their distinct tasks As such a continual dialog is necessary To accomplish this there are advisors from both disciplines on the project and representatives from the Systems team are required www A TIBOOK ir A Systems Engineering Approach to the Development of an Autonomous Sailing Vessel 95 to attend meetings of the Naval Architects Further during a Fall semester course offered to the Naval Architects the Systems advisor gives an overview of the ba sics of the automation and control of the previous year s vessel and explains some of the common issues related to the systems and naval architecture interaction Planning To execute a project of this nature in only one semester takes
34. annually since 2006 except 2007 SailBot is open for semi autonomous and fully autonomous sailing boats up to 2 m in length Since 2010 they provide an additional open class for boats with a maximum length of 4 m For the 2011 competition five different events are an nounced 1 fleet racing 2 station keeping 3 endurance contest 4 autonomous navigation and 5 presentation and design Mail from Colin Sauz on Microtransat mailing list from 02 10 2010 10 Web site of SailBot 2011 http www sname org SNAME SailBot2011 Home Default aspx www A TIBOOK ir 12 R Stelzer and K Jafarmadar 3 2 3 World Robotic Sailing Championship and International Robotic Sailing Conference The World Robotic Sailing Championship WRSCJ is an international competi tion for autonomous sailing boats The first WRSC was held in Austria in 2008 Since then it took place every year Portugal 2009 Canada 2010 and Germany 2011 The competition is open to boats of up to 4 m in length Detailed rules of WRSC change every year to respond to recent scientific developments and stimulate certain areas of research By keeping a rather low entry threshold WRSC not only appeals to experts but also provides a platform for new teams in this recent field of research The competitions coincide with the International Robotic Sailing Conference IRSC This conference is the basis of scientific exchange in the robotic sailing community The combination of
35. apparent wind angle the GPS speed and apparent wind angle were sampled with 5Hz and the minimum speed as well as mean and standard deviation of the apparent wind angle were computed for a sliding window of 50 samples The data was stratified by the apparent wind angle and restricted to episodes where the minimum speed was larger than 0 5m s The actual experiments where done using four almost identical boats and sailing in moderate conditions of approximately 3 Bft Figure 8 summarizes the max min GPS speed m s all max min GPS speed m s all boats window 10s SD lt 10 deg boats window 10s SD lt 30 deg 20 0 340 20 340 40 320 40 320 140 220 160 180 200 Fig 8 An illustration of the boats sailing performance in winds of approximately 3 Bft as polar plots The plots show the maximum of the minimum speed in a 10s window for each apparent wind angle For the left plot the standard deviation of the apparent wind angle was restricted to be smaller then 10 i e episodes with a larger standard deviation where excluded The right plot relaxes the latter requirement to 30 indicating that it is particularly hard to maintain a constant apparent wind angle on a downwind course www ATIBOOR ir A New Class for Robotic Sailing The Robotic Racing Micro Magic 81 speed over ground yellow speed over ground green 1000 1500 1500 2000 Frequency Frequency 1000 500 500 oO oO 0 0 0 2 0 4 0 6 0 8 0 0 0 2 0 4 0 6
36. by lines Nowadays the wind vane self steering devices are also used on sailboats We remarked that those devices were put on the stern In fact a boat can often be controlled single handedly by a human so naturally the wind vane the rudder sail lines are on the stern Besides the bow is often hardly accessible Finally the bow is subject to waves splash and more rough conditions On the other hand robot actuators can be dispatched everywhere on its body so that is when we came with the idea to put the self steering system on the bow Figure I illustrate the difference between www A TIBOOK ir Sailing without Wind Sensor and Other Hardware and Software Innovations 29 regulators sail a Fig 1 Sub Figure b shows a classical design of a wind vane self steering device when the rudder is on the stern In this case there is a need to invert the rotation between the wind vane and the rudder using gears for example Sub Figure a shows the new simplified design where the rudder is on the front and the inversion of rotation is no longer necessary the bow and stern designs The main difference is that there is no need for gears to invert the rotation between the wind vane and the rudder We will remark later in the simulation part that those two designs are equivalent in terms of trajectory stabilization 2 2 Simulation 2 2 1 Equations Consider a dynamic system defined by the following evolution function x f x u 1
37. constructed in late 2004 as a proof of concept for a small but durable sailing robot The hull is about 1 5 m long and is rigged with a 1 m high wing sail 34 ARC Figure is about 1 5 m in length and features two independently controlled wing sails and two rudders controlled by a single actuator It is equipped with a gimbaled compass GPS receiver and a combination of an AtMega128 mi crocotroller and a Gumstix24 single board computer running Linux The only power source is a bank of 20 pieces of 1 2 V AA size recharchable batteries with a capacity of 2500 mAh each 40 Beagle B Figure 8 dif is their largest boat and was constructed in late 2006 by Robosoft a French robotics company It is 3 5 m long and uses a 3 m solid wing sail Beagle B is intended to provide a serious oceanography platform for long term missions Its power is provided by two 15 W solar panels and four 60 Ah batteries with 12 V It includes a YSI 660 Sonde for gathering oceanographic data as well as an Iridium SBD transceiver and GSM modem for data transmission Beagle B participated in the Microtransat Challenge 2007 in which it sailed a total of 25 km over 19 hours 21 Paul Miller 22 Paul Miller 23 Paul Miller 24 Colin Sauz 25 INNOC 26 http www gumstix com 7 Colin Sauz www A TIBOOK ir 18 R Stelzer and K Jafarmadar Pinta Figure 8clp was built for WRSC 2008 and the Microtransat transatlantic race Unlike the other boats
38. controller needs to provide a sys tem tick a periodical interrupt in this case with a period of 1 ms as a basic clock for the scheduler The LM3S6965 has a special timer to provide the system ticks 4 1 2 Board Support Package To hide more of the hardware details a board support package BSP provided as a part of the port of uC OS II for the Cortex M3 is used The BSP may be extended corresponding to the systems requirements Figure 6 shows the initialization of the AS5040 to get the wind direction Commands with a leading BSP_ are special com mands of the uC OS U BSP while the other commands are part of the driver lib 4 1 3 Tasks To create a task uC OS II needs some initialization steps see Fig 7 After BSP_Init CPU_Init and OS_Init the main task is created In this task all needed user tasks are created Then the OS may be started line 12 After this step the multitasking environment is running 4 2 Interrupt Service Routine vs Task To show the different handling e g to get sensor data we compare the wind di rection using the AS5040 with the heading using the compass see Fig 8 The wind direction is a PWM signal which needs to be analyzed by the controller Af ter the initialization an interrupt service routine ISR is called whenever the logic level on the input pin changes As result we got two values see Fig 9 The whole www A TIBOOK ir Using ARM7 and uC OS IH to Control an Autonomous Sailboat 109
39. direction is between 0 and 13 de grees we do not adjust the rudder if the angle is between 13 and 28 we turn the rudder of 10 degrees between 28 and 55 degrees we turn of 20 degrees between 55 and 80 degrees we turn of 30 degrees and finally if the angle is bigger than 80 degrees we turn the rudder of 45 degrees We prest the same strategy for the sails angle as shown in Fig 8 4 Development of Research Programs around Breizh Spirit A research program is in current development to provide experimental data to vali date the seakeeping codes that are realised at LBMS ENSTA Bretagne This program consists of realising a measurement system for recording the move ment of a boat at swell The interest of using a sailing robot is first off all its small size Thus sailing in a usual seastate condition gives situations where the waves become big compared to the boat dimensions www ATIBOOR ir 66 R Leloup et al The second interest is the control we have on each actuator of the boat This is very important because with classical sailing boat tests there are big uncertainties concerning the crew behaviour the positioning and the trimming of sails 8 Of course it is possible to get information from video recording But it complicates and delays the analysis of the experimental data recording Sailing robots have the ad vantage that all parameters can be easily known at each time step of the experiment 5 Tests and Results The
40. environment the HIL simulator and BeagleB A triangular course of approxi mately 1 2km was selected This was based upon a race undertaken on July 11th 2009 during the 2009 WRSC World Robotic Sailing Championships in Matosin hos Portugal The course was sailed in both simulated environments and compared with the actual data from BeagleB during the 2009 WRSC In all cases the wind was blowing from the North West at approximately a Beaufort force 2 Table 2 compares the time taken actual distance covered and number of actuator movements made by each Figure 4 compares the courses of all three From Figure 4 it can be seen that the courses are not quite identical There is a slight offset between them caused by different co ordinate systems used by Tracksail and the hardware systems the starting point also varied slightly due to an initial www A TIBOOK ir Simulating Sailing Robots 121 BeagleB EN HIL Simulation Tracksail Center 41 17292 8 69813 200 ft Map created at GPSVisualizer co m 100 m 1 Map data from OpenStreetMap org Fig 4 Map showing the GPS tracks Table 2 A comparison between the simulator HIL simulation and BeagleB Type Distance Sailed Time Taken Rudder Movements Sail Movements Simulator 1 69 km 7403 seconds 196 33 HIL Simulation 1 98 km 902 seconds 138 528 BeagleB 2 32 km 5168 seconds 267 96 drift while the control system was starting A bug in BeagleB s code caused it to str
41. for Robotic Sailing The Robotic Racing Micro Magic 71 50 Hz Furthermore an on chip algorithm computes tilt compensated compass data with azimuth and roll pitch resolutions of one degree The ITG 3200 is a three axis MEMS gyroscope providing calibrated rotational measurements with a resolution of 1 14 degrees at measurement frequencies of up to 2000 Hz The wind speed and wind direction sensors are custom made based on the popular AS5040 sensor Austria MicroSystems Austria Finally for absolute position and speed over ground we used the ublox LEA4 T GPS chipset ublox Switzerland mounted on a custom designed board The GPS unit connects to the main controller via a serial interface using the binary UBX protocol 4 Control As indicated before the onboard microcontroller is mainly reading and pre processing sensor data setting the servo positions and running the communi cation Figure 5Jillustrates that the current board operating at 50 Hz for the internal sensor processing loop and 5 Hz for the external communication loop has a mod erate duty cycle The limiting factor would be an increase in the frequency to send data via bluetooth The main control runs onshore primarily to have sufficient computing power to allow studying more sophisticated methods e g for path planning and collision avoidance While basic processing of sensor data and control of sail and rudder po sition is done at 5 Hz other tasks like filtering sensor da
42. image and set the pixel values the following way www A TIBOOK ir 132 O Gal 1 For each pixel in the learning zone check if this pixel is inside the search space 2 If so set pixel x y value to k 3 Choose the next pixel according to f value The occurrence matrix allows us to find similar texture inside the image The learning zone can vary from image to image as shown in Figure B For an n m image we use 10 percent of n on each side left and right and 10 percent of m at the bottom as learning zone Fig 3 Learning Sea Pattern on the left we can see the original image and the right side shows the image after applying occurrence matrix 3 4 Morphologic Cleaning After completing the detection process of the sea pattern there are still few sea pix els that were not detected We use a unique cleaning filters to improve and smoothen the image so target identification will be easier The first filter is the Fill Filter This filter fills the target with missing pixels 1 e for each black pixel sea pixel We scan the eight neighbors of each pixel If the pixel s eight neighbors are white target pixels we change the pixel value to be a target pixel www A TIBOOK ir Automatic Obstacle Detection for USV s Navigation Using Vision Sensors 133 We use the Fill Filter Matrix Fy 111 F li j 101 9 111 The second filter is called Cleaning Filter This filter makes exactly the oppo site action from t
43. is only one target colored in red Fig 9 Labeling Algorithm Matrix n x n matrix for each target each cell in the matrix i j defines the connected component related to the interesting point in this case we classify only one target with serial numbers of 1 and O 3 7 Multiple Targets One of the main goals of the algorithm is the ability to track multiple targets at the same time This can be done by classifying the interesting points Classifica tion of the interesting points is based on connected components using labeling algo rithm The input image for the labeling algorithm is a target image after morphologic cleaning In case of multiple targets the labeling algorithm recognize and label each www A TIBOOK ir 138 O Gal component in a different serializing number Then each interesting point s class fied according to its matching number Example of labeling algorithm matrix n x n can be seen in Figure 8 Each cell in the matrix i j defines the connected compo nent related to the interesting point This method enables us to track multiple targets in an efficient way Figure 9 shows the output of the labeling algorithm Each label is colored with a different color in this case a single component colored in red 4 Results Our algorithm was tested with five hours of video stream from a database with real video recorded with a camera on a USV in different light and sea states The algorithm was tested on real time vi
44. is pointed to the literature 4113 A central notion in QSR is that of a qualitative calculus which comprises a finite set of atomic relations to describe the relationships between entities as well as oper ations on these relations Technically these relations are binary sometimes ternary relations between domain level objects In this paper we only consider binary re lations Relations feature a set theoretic semantics i e a binary relation r on the domain D is a subset r C D x D i e a set of ordered pairs x y with x y D The set R of all relations of a qualitative calculus is assumed to be jointly exhaustive and pairwise disjoint Hence we have a boolean set algebra with the usual set op erations Elements of this set algebra i e arbitrary disjunctions of atomic relations are also called gualitative relations for short Qualitative calculi define two relation operations which allow new facts to be derived from given ones conversion and composition Mathematically qualitative calculi relate to relation algebras in the sense of Tarski Conversion can be inter preted as shifting perspective from one entity to another For example conversion allows us to infer from the fact that L beck is NortEast of Bremen that it also holds that Bremen is SouthWest of L beck The composition operation allows knowledge about a common entity to be combined Only using that Hamburg is NortEast of Bremen and L beck is NortEast of Hamburg we can
45. known as Metal Mick as it was capturing much of the behaviour of an experienced helmsman It compensated for varying sea states using feedback control and automatic gain adjustments This lead to a first simple adaptive autopilot The work of Minorsky is also regarded as making key contributions to au tomatic ship steering Nicholas Minorsky presented a detailed analysis of a position 3 Apparent wind is referred to as the velocity of air as measured from a moving object as a ship 4 Point of sail describes the direction of a boat with respect to the direction of the wind www A TIBOOK ir 6 R Stelzer and K Jafarmadar feedback control He formulated the specification of a three term controller better known as proportional integral derivative PID controller 2 1 3 Intelligent Rudder Control Conventional electronic self steering systems found on the majority of vessels at sea still employ PID control algorithms to control the heading 18 Van Amerongen 52 identified two major disadvantages of this type of controller 1 It is difficult to adjust manually because the operator usually lacks the insight into control theory 2 The optimal adjustment varies and is not known by the user Changing circum stances require manual readjustment of a series of settings Due to the highly dynamic and ever changing environment artificial intelligence AI techniques like fuzzy logic FL artificial neural networks ANN and combi na
46. necessity of a full re routing once new forecasts are available 4 Conclusion and Future Work We proposed a long term routing algorithm which finds an arbitrarily accurate ap proximation to the optimal route for a sailboat for real life wind conditions Our approach takes changing weather conditions into account by dynamically adapt ing the underlying graph used as input to the A algorithm It is is implemented in Constraint Handling Rules and compares very well to existing solutions To our knowledge it is the first published implementation of an long term weather routing algorithm for sailboats in a declarative or rule based programming language www A TIBOOK ir A Rule Based Approach to Long Term Routing for Autonomous Sailboats 203 4 1 Future Work We use deterministic weather forecasts which are only available for a certain time ahead see Section 2 1 1 So called ensemble forecasts consist of different scenarios that can happen with given probabilities and are usually available for a longer time ahead Research for sailing yacht races has shown how ensemble forecasts can be used to find optimal routes which perform well under all possible scenarios 7 An extension of the routing algorithm to handle ensemble forecasts could make the routing more realistic and would allow for accurate planning of even longer routes Besides wind conditions the travel time of the sailboat can be affected by cur rents as well Incorporating ocean c
47. not created until two adjacent nodes are present DOO Roboat Router map wind vectors waypoints Forecast 9 2010 08 12 02 00 GMT Ferrara 54 HLoHe 1 2 Dr CUT Start Lon 5 30 Lat 49 60 Date 17 03 2010 Gi Time 0200 GMT Dest Lon 4 40 Last 3h HR Caklate route Fig 3 The GUI showing wind conditions and the calculated route www ATIBOOR ir 200 J Langbein R Stelzer and T Fr hwirth e Ifa node is added to the closed list every incoming and outgoing edge to and from this node can be removed from the graph as nodes in the closed list will not be considered again The dynamic construction of the graph also facilitates another optimization When a node x is expanded the time tstart C X at which this node will be reached by the boat is known Consequently the forecast in the GRIB file valid at this point in time is known and the calculation of the travel time for outgoing edges of x has to be done only for this forecast and not for all forecasts in the GRIB file 3 Implementation in CHR We implemented our algorithm mainly in Constraint Handling Rules CHR com bined with SWI Prolog For an introduction to CHR we refer to 4 Our imple mentation is based on an existing implementation of Dijkstra s algorithm with a Fibonacci heap in CHR 10 which is used as open list to achieve optimal time complexity and was extended to incorporated the use of the heuristic function Our
48. of the rudder assembly which had proved difficult and time consuming Fig 5 The rudder assembly magnetic linkage Fig 6 A photograph of MOOP2 sailing in the and servo sea off Aberystwyth www A TIBOOK ir MOOP A Miniature Sailing Robot Platform 47 Table 3 The specification of MOOP3 Name MOOP3 Date of Construction Summer 2009 added a Gumstix and rotary wind sensor in June 2010 Sails Twin Wing Sail Sail Actuator Modified 360 degree servos Rudder Actuator Servo on magnetic linkage Computers PIC18LF4550 Gumstix added in June 2010 Wind Sensor Initially no wind sensor rotary sensor added in June 2010 Batteries 5 13 Ah NiMH rechargeable size F batteries Notes Competed in the 2010 WRSC to construct During this process the possibility of also removing the wind sensor was suggested As one of the most vulnerable components in the boat this would eliminate a major potential point of failure By setting the sails for a given point of sail it was hoped that the boat would settle upon a course and that by observing the compass heading and roll of the boat that it would be possible to determine when the boat had settled down Now that it was known which direction the boat was trav elling for a given point of sail the wind direction could be derived Since originally proposing this method Xiao et al have devised and demonstrated alternative method based upon a Voronoi diagram this work suggests that it is quite possi ble to operate
49. on deck using a carbon rod to connect the servos with the boom fitting see Figure I On the one hand advantages include minimal backlash direct sail position control via the servos and the ability to back the sails On the other hand the servos must be sealed against water and the axis of rotation for the sails must be considered While the main sail rotates about the vertical mast the jib rotates about the inclined forestay Hence the horizontal rotation of the servo must be converted into a rotation about the stay as illustrated in Figure 2 Note that we have added a horizontal offset as shown in Figure Pb and use the ball bearing position to control how much the boom can rise see Figures 2b and 2H www A TIBOOK ir A New Class for Robotic Sailing The Robotic Racing Micro Magic 75 A IM N Fig 2 The figure illustrates how the rotation from the deck mounted servos is converted into a rotation about the forestay a both axis are shown and the triangle ABC has to rotate about its side AB b the actual design adds an offset betwen forestay and axis of rotation but the triangle is still rotating almost parallel to the forestay c in closed position the boom is low and keeps the leech tight and d when the sail is open the boom rises to keep the leech tension Note that by adjusting the position of point B the leech tension in the open position can be controlled Given the already relatively small sail area it is rather untypical
50. on ocean monitoring But some more tasks are possible to be fulfilled by manned or unmanned sailing robots Intelligent sensor buoys CO gt neutral transportation of goods Reconnaissance and Surveillance Supply Vessel Unmanned ferrying Minefield mapping Acknowledgements This paper was written as part of AAS Endurance a joint research project of INNOC Austria and Oregon State University USA The aim of the project is to develop an autonomous sailing boat for passive acoustic monitoring of marine mammals and mitigation of human impacts on them The project is realised within the funding programme Sparkling Science supported by the Austrian Federal Ministry of Science and Research References 1 Abril J Salom J Calvo O Fuzzy control of a sailboat International Journal of Approximate Reasoning 16 3 4 359 375 1997 http mapp1 de unifi it persone Allotta ICAD Abri1l1997 pdf 2 Adriaans P W From knowledge based to skill based systems Sailing as a machine learning challenge In Lavra N Gamberger D Todorovski L Blockeel H eds PKDD 2003 LNCS LNAI vol 2838 pp 1 8 Springer Heidelberg 2003 http www springerlink com content 9f1801mf9pkwdgdy 3 Allensworth T A short history of Sperry Marine 1999 http www sperrymarine northropgrumman com Company Information Corporate History Sperry History 4 von Alt C Autonomous underwater vehicles In Autonomous Underwater Lagrangian Platfor
51. or the Austrian Roboat 7 As another alternative model sailboats have been used and are often eas ier to handle and comparatively low priced We have previously described our ex perience with very small one design boats for robotic sailboat racing 2 Based on our initial results we have refined our design based on the popular Graupner Micro Magic kit We propose the robotic racing Micro Magic rrMM as a new class at the World Robotic Sailing Championship This paper presents the motivation for establishing the new class and discusses the proposed class rules Moreover we summarize the hard and software design to enable other teams building an rrMM class boat Finally we give initial results with respect to the sailing performance and the overall handling of the boat 2 Class Rules Our key motivation to study one design boats was to establish a testbed for algorithm development i e the sailing performance should mostly depend on the methods used to control the boat A further objective was a small size so that the boat can be easily carried and launched even on smaller lakes Moreover most parts should be readily available and simple to assemble as the focus is on control algorithms and not on naval design Furthermore the cost for parts and components should be low to encourage a wide adoption of the class Finally the boat should be sailable in a sufficiently wide range of conditions to allow frequent testing without depending to
52. performance problems handling www ATIBOOR ir Collision Avoidance Using a World Server 165 Table 2 Network traffic generated by the World Server with different number of connected boats clients and buoys The boats were simulated by the communication gateway and each sent 255 Byte s to the World Server We used multiple instances of our GUI to simulate the clients each instance polled the World Server two times per second Boats Buoys Clients Downstream kB s Upstream kB s 10 15 1 5 7 10 15 5 6 3 25 10 15 10 8 2 48 20 15 10 13 8 85 20 15 20 18 5 160 a large number of boats and clients as the traffic stays low Additionally our test have shown that CPU load is no limiting factor as it always stayed below 15 even with more than 100 objects 6 Conclusion By introducing the World Server we have presented a data storage system that en ables all participants in a race to have access to a global view of the race situation This system can be seen as a basis for further developments in autonomous sailing Due to global data access for all clients it is easily possible to implement collision avoidance without active sensors Furthermore a central judging and race control ling system can be developed In addition the logging of data during a race generates a database of authentic boat movements that e g allows to test new algorithms by replaying the records in a simulator References 1 The microtransat http www mic
53. position oscillated between two posi tions the control system only has 11 possible positions it can place the sail actuator into for several minutes This problem has been observed in both BeagleB and Tracksail but on this occasion only happened to occur with the HIL simulation Despite these differences the performance between all three systems is broadly similar with differences being within an order of magnitude apart This suggests that using either Tracksail or the HIL simulation are likely to see similar results to those of BeagleB When using these methods to verify algorithms and experiments to run upon BeagleB they should therefore be sufficient to at least eliminate those which will not be viable Power management results are not available at this stage as the version of the ex periment on BeagleB was performed before power logging hardware was installed The Arduino and current transducer system has also been problematic in initial de velopment due to random reboots of the Arduino under high current loads and ac curacy issues with the current transducer under low loads A crude comparison of power useage can be made by comparing the number of actuator movements Given that these are of similar orders of magnitude this indicates that there is a reasonable match between the simulator HIL and BeagleB in terms of actuator use and actu ator power consumption However these values do not give a complete picture as they do not distinguis
54. reasoning This paper demonstrates the utility of qualitative reasoning in autonomous sea navigation In previous work we have studied how purely symbolic reasoning can help to consistently integrate pair wise rule constraints when multiple agents meet 6 We now focus on the problem of actually controlling a vessel in a rule compliant manner The contribution of our work is to show how the official right of way rules for vessel navigation COLREGS vessels in sight of each other according to the International Maritime Organization IMO can be modeled using qualita tive spatial representation Furthermore we show how the representation supports rule compliant action planning for autonomous vessels This paper is organized as follows We start by putting our approach in the context of high level agent con trol SectionB introduces qualitative representation and reasoning techniques Using these techniques Section 4 details our formalization of navigation rules Section 5 explains how we incorporate the qualitative rules into action planning We give an experimental account of our approach in Section 6land conclude the paper by sum marizing the results and discussing further research directions 2 Rule Compliant Navigation Rule compliant navigation starts by formalizing navigation rules in a formal language that can be understood by autonomous agents To this end symbolic nav igation rules are suitable 11 in particular qualitative represen
55. route Sail from 0 0 to 110 100 on a bend route restricted in width to 20m with wind from an arbitrary direction 3 Giving way to an oncoming vessel rule depicted in Fig 5 a Sail from 0 0 to 100 0 with wind from a random northern direction compass angle between 270 and 90 avoiding an oncoming vessel 4 Crossing a frequented channel rule depicted in Fig 5 b Sail from 0 50 to 100 50 with wind from the west compass angle 225 and 315 passing behind the stern of two crossing vessels The challenge is to start sailing slowly which is not favored by the heuristic in order to pass behind the other vessels before increasing sailing speed The vessel always starts with zero speed and the wind speed is 3ms For a fixed amount of n active nodes we determine whether the planner is able to determine a solution and we record the execution time of the plan as well as the length of the trajectory computed We regard planning as successful if the planner can determine corresponding to a trajectory to a position closer than 10m to the goal the planner www A TIBOOK ir 152 D Wolter F Dylla and A Kreutzmann 1 sailing along a route 2 following a narrow route 7 N A o ea 3 giving way to an oncoming vessel 4 passing a channel
56. scan lines after a power on may contain wrong angle infor mation This resolves within one full turn of the transmitter 6 2 2 Register 03 Zoom Level Payload length 4 bytes 1 dword The register 0x03 controls the range of radar operation The scan radius is en coded as dword in decimetres The format suggests that any value within a valid Table 2 Example of packet payload data for setting a register 0 1 2 3 4 5 03 C1 10 27 OO 00 Set range to 1 km reg cmd data reg Register number It describes the function to access cmd Always OxC1 for write commands data variable length payload to be written at least one byte Multi byte fields are encoded little endian www A TIBOOK ir 176 A Dabrowski S Busch and R Stelzer range can be set We suggest to stick to the predefined ranges see table 3 as used by the manufacturers display unit as we can not confirm if other values produce accurate data A change is reflected in the scan line header in the radar image data stream in typically under a second 6 2 3 Register 06 Filters and Preprocessing Payload length 1 byte selector 8 byte data 9 bytes Register 0x06 offers access to a wide range of filters and preprocessing functions such as Sea Clutter Compensation Rain Clutter and Gain control The first byte of the payload acts as selector a comprehensive description can be found in table 4 Note There is no automatic rain clutter adjustment x2 is in the range fr
57. still derive with composi tion that L beck is NorthEast of Bremen Since QSR is involved with finite sets of atomic relations only the composition operation is usually provided in form of look up tables called composition tables These operations are particularly impor tant for constraint based qualitative reasoning and often enable efficient algo r thms to tackle the problem of deciding whether a set of constraints involving qualitative relations is consistent or not Conceptual neighborhood extends static qualitative representations by interrelat ing the discrete set of base relations 7 Two spatial relations of a qualitative spa tial calculus are conceptually neighbored if they can be continuously transformed into each other without resulting in a third relation in between We note concep tual neighborhood on the qualitative level corresponds to continuity on the physical level For example let us consider the relations behind same and ahead to relate the positions of two vessels in a match race For reasons of simplicity we assume that vessels are only able to move forward with changing speed In the leftmost con figuration shown in Fig 3 vessel A is behind B Observing the scene a few minutes later shows that now A is ahead of B Assuming continuous motion it is not possi ble for A to overtake B without passing B at some time 1 e being at the same level Therefore ahead and behind are not conceptually neighbored whereas ahead an
58. subject to regulations For example right of way regulations govern how to travel public spaces Action planning for an autonomous agent needs to respect right of way reg ulations These rules are special in that they have been designed for the general public and are denoted in natural language using abstract concepts of space Mak ing these regulations accessible to an artificial agent requires translating them into a formal language that can be understood by the agent and which seamlessly inte grates with the agent s navigation process In order to facilitate correctness and ver ifiability of the translation an abstract language is particularly suited if it is able to reflect the concepts originally used in the right of way regulations We use qualita tive representations to abstract real wold observations to abstract knowledge about Diedrich Wolter Frank Dylla Arne Kreutzmann SFB TR 8 Spatial Cognition University of Bremen e mail dwolter dylla kreutzma sbftr8 uni bremen de www A TIBOOK ir 142 D Wolter F Dylla and A Kreutzmann space and time on a conceptual level Qualitative spatial representations see for an overview aim to provide a formal model for human level commonsense un derstanding of space and time Moreover they enable abstract reasoning processes Technically qualitative representations summarize similar real world states by a dis crete finite set of qualitative categories that give rise to symbolic
59. system To test the longevity of components and ability to sail at sea one proposed experiment is to drop an MOOP without a control system or only a very simple heading holding system into the sea and monitor its progress via a satellite tracker Acknowledgements The authors would like to thank Barry Thomas and Tom Blanchard for their contributions towards designing constructing testing and repairing the MOOPs References 1 Spot satellite messenger accessed June 10 2011 2 Dual wing sailed moop sailing robot 2009 accessed June 10 2011 3 Twin wing miniature sailing robot 2010 accessed June 10 2011 4 Ammann N Hartmann F Jauer P Bruder R Schlaefer A Design for a robotic sailboat for wrsc sailbot In Proceedings of the 3rd International Robotic Sailing Con ference Kingston Ontario Canada pp 41 43 June 2010 5 Ammann N Biemann R Hartmann F Hauft C Heinecke I Jauer P Kruger J Meyer T Bruder R Schlaefer A Towards autonomous one design sailboat racing navigation communication and collision avoidance In Proceedings of the 3rd Interna tional Robotic Sailing Conference Kingston Ontario Canada pp 44 48 June 2010 www A TIBOOK ir MOOP A Miniature Sailing Robot Platform 53 6 Benatar N Qadir O Owen J Baxter P Neal M P controller as an expert system for manoeuvring rudderless sail boats In UK Workshop on Computational Intelligence UKCI 2009
60. the localhost using the default port WorldServerClient client new WorldServerClient 127 0 0 1 Try to connect if client connect Create a SailBoat object representing our boat SailBoat boat new SailBoat myBoat WinningTeam set some data boat setLatitude 53 5 boat setLongitude 10 0 boat setHeading 180 0 boat setSpeed 2 5 send data to the server client updateWorldServerData boat retrieve data from the server IDictionary lt String WorldServerRecord gt data client get WorldServerData do something with the data disconnect client disconnect 4 3 Communication Gateway In our setup client and boat communication are split into two seperate modules The communication gateway shown in Fig 2 is conceptually not directly associated with the World Server but abstracts from the details of Bluetooth It can be configured to forward all data received from boats directly to the world server so the world server gets the data no matter if a client uses the World Server or not 5 Results The World Server was tested in simulations as well as under real race conditions The system was successfully used in a fleet race serving four boats multiple buoys and obstacles as well as up to five clients one for each boat and one to visualize the world state Table 2 shows results for simulations with different numbers of virtual objects and clients As illustrated the World Server has no
61. the mast Wires run from the computer through the mast to a female DIN plug A tube in the centre of the sail fits over the mast and located inside it is a male DIN plug wires then run from this plug to a wind sensor a the top of the sail See Figure l for a diagram of the sail assembly The sail is constructed by cutting a wing shape from a block of polystyrene using a hot wire cutter hollowing out the centre of each to make room for a carbon fibre shaft which slides over the mast and the wires which connect to the wind sensor The hollowed out area is then filled with polystyrene balls and epoxy resin Finally the sail is covered in glass fibre cloth and painted with epoxy resin to give it rigidity A combination of lead shot and epoxy resin fills the keel to provide ballast This also allows for the outer skin of the keel to be pierced without water being able to www A TIBOOK ir MOOP A Miniature Sailing Robot Platform 41 enter the boat Figure I shows the dimensions of all components in MOOPI the second MOOP to be built and Figure 2 shows the internal electronics and compo nent layout of MOOPI1 The rest of this section describes the differences between each variant of the MOOPs and the reasoning behind those differences 2 1 MOOP0 and MOOPI The first boat to be constructed was known as MOOPO and was begun in late 2008 Its hull design differs slightly from the others due to the way it was constructed A series of 15 cross sectio
62. their design projects 1 Through the remainder of this section we will discuss an appropriate scope for the SailBot project and lay out the objectives and performance metrics For purposes of this discussion we will assume that the problem and needs state ments for the design are completely defined by the fixed vessel provided by the Naval Architecture students and the SailBot competition rules 2 3 Further there are constraints on weight and on size volume of each component that come from the fixed vessel design and must be discussed with the Naval Architecture students It is impossible for example to place a standard laptop PC into the Sail Bot designs that have thus far been generated at USNA due to the access hatch size and hull form so selecting such a system is a priori prohibited 1 2 1 Objectives and Functions The first step in a proper system design after a clear needs analysis is to lay out the objectives of the design An objective for a design is one of a set of character istics that differentiate a good design from a bad one 1 Objectives describe what the system is not what it does and a fully defined objectives tree is a crucial step toward developing a good project that meets the needs of the problem Students typically struggle to generate the appropriate depth and breadth of ob jectives for a complex project Wanting to focus on what the vessel does func tions and how it will accomplish those functions mea
63. to adjust the sails to the wind force i e to reef Instead the keel can be changed to a heavier bulb which can extend the sailable wind range to up to 5 Bft Figure 3 shows the standard 370g bulb from the kit and two performace bulbs RT Modellbau Germany with 370g and 470g respectively Note that the keel position can be moved slightly for ward and aft to trim the boat but generally it is advisable to mount the heavier keels further towards the stern 3 2 Electronics The on board electronics is based on an ATmega 2560V Atmel CA microcon troller operating at 11 0592 MHz The chip features built in hardware for jitter free phase and frequency correct waveform generation which we use to generate pulse www A TIBOOK ir 76 A Schlaefer et al 200 150 round trip time ms 100 50 0 0 100 200 300 400 distance m Fig 3 Three different keel bulbs a the Fig 4 The round trip data transmission standard 370g bulb b a 370 g performance time from boat to onshore computer indi bulb and c a 470g performance bulb cating that the latency is below 20ms for typical operating distances of the boat width modulation PWM signals for the three RC servos moving rudder main sail and jib Likewise the microcontroller is also reading and preprocessing the sensor data and controlling communication with the onshore computer For data transmission we use long range bluetooth modules F2MO3GXA SOI Free2Move S
64. to be used as sample data for steering and obstacle avoidance algorithms In figures 6 7 comparisons of photo and radar data give an idea of how well the setup might detect objects in fine weather conditions 8 Power Consumption The electronics in the radar antenna unit has no power switch It gets supplied and boots up as soon as there is power at the radar interface box but does not turn on the transmitter In this standby mode the unit consumes 0 15 A at 12 V In full operation Fig 6 Radar data photo comparison Alte Donau with canoeist radar port side www A TIBOOK ir A Digital Interface for a Lowrance Broadband Radar 179 Fig 7 Radar data photo comparison Alte Donau with buoys radar port side the unit consumes about 1 4 A The display unit has a power switch and consumes about 0 7 A when turned on Values may vary depending on mode of operation or settings 9 Sample Implementation Simultaneously with the paper an open source sample implementation in JAVA is released and available athttp www roboat at technologie radar It provides the following features Source selection operate on live radar network connection or on a pre recorded PCAP Fild Activation turn the transceiver on and off Range selection set the zoom level which is also reflected in acquisition parame ters Image decode image data is decoded and displayed in Cartesian form Filter settings may be adjusted over the network 10 Data
65. u K x det uag ug K2 e 13 In practice there are some conditions so that this formula works better For ex ample we started by regulating the angle of the boat first before caring about the distance e Ka 0 while det uag ug important For more stability when the boat is nearly on the line AB we replace K xe by Ko arctan K3 xe When it is possible to have a good estimation of the speed vector which is not always the case especially if the speed of the robot is slow we propose a com mand giving similar results to the last hybrid command The idea is to regulate the boat such as its speed vector denoted by m points the waypoint B The command becomes u K x det uyp Uy 14 Note that because of the drift u ug www ATIBOOR ir 36 J Sliwka et al Fig 10 The robot L improbable 4 Our Prototype 4 1 Introduction In 2011 two prototypes were developed at ENSTA Bretagne One is a copy of Breizh Spirit and the other is L improbable see Figure 10 which is based on the Optimist type boat design http optiworld org L improbable is designed to try new solutions in terms of long period ocean navigation 4 2 Mechanical Design of L Improbable The core of the boat is an aluminum board which acts as a keel and supports the mast as seen in Figure 7 The rudder is attached to the plaque as well Surrounding the board there is an extruded polystyrene made hull The hull was profile
66. v Wi Oj Vv W Q As sailing directly upwind is not possible sometimes it is required to beat in order to sail from x to x We incorporate this into the travel time calculation by approx imating the boats velocity made good along the great circle path between x and x in the following way As described in 7 we neglect the time for tacking and compute the velocity made good by using the convex hull of the polar diagram for speed calculations when sailing upwind The true wind angle at which the boat is required to beat can be configured by the user www A TIBOOK ir 198 J Langbein R Stelzer and T Fr hwirth Fig 2 An exemplary part of the routing graph Wind vectors for each node are shown as bold arrows edges to and from the center node as solid lines Edges connecting the surrounding nodes are denoted as dashed lines 2 1 3 Routing Graph Oceans constitute a continuous search space with an infinite number of possible waypoints To reduce the search space and make classical shortest path methods applicable to the routing problem we chose to discretize the search space into a grid graph with equidistant nodes representing points on the sea Each node is connected to its eight nearest neighbors by directed edges Nodes located on a land mass which are detected using a binarized world map are not included in the graph Nodes at locations with hazardous wind conditions that is to say locations at which the wind speed is
67. wide sector immediately behind the boat This avoided downwind stability problems previously found in MOOP3 MOOPn was taken for an acceptance test sail on Llyn Yr Orefa and successfully completed a triangular course Although a single wing MOOP was not used for comparison at the time the performance of MOOPn was a noticeable improvement There were none of the violent jibes seen by MOOPO and MOOPI when sailing downwind Up wind performance also appeared more stable with no accidental tacks being noted A video from this test is available from 3 3 2 Sea Tests In addition to lake tests several sea and large lake tests have also taken place with various MOOPs 3 3 WRSC 2009 The first sea test took place in the Atlantic off the coast of Matosinhos Portugal during the World Robotic Sailing Championships in July 2009 A triangular course www A TIBOOK ir MOOP A Miniature Sailing Robot Platform 51 of approximately 1 2 km in length was set The first two portions of this were on a broad reach and the final was straight upwind MOOPO was used and successfully sailed the first two portions of the course but then failed to make the turn onto the third This was believed to be due to the tidal current which was moving in the same direction as the wind The boat was then carried upwind and allowed to sail back to the second waypoint with the wind now straight behind the boat It made progress towards this waypoint but jibed almost constantl
68. y Ix y 1 Ix y 1 5 3 Gy x sY where fac in Eq I sets the smoothing quality The Adaptive Smooth filter contains two Gaussian matrixes Cleaning and Blur ring matrix calculated by Eq I The Cleaning Matrix Cli j is an N x N matrix in our case N 7 the specific values that were found to be the most efficient in the simulations are detailed below 0 04 0 04 0 04 0 04 0 04 0 04 0 04 0 02 0 02 0 02 0 02 0 02 0 02 0 02 0 00 0 00 0 00 0 00 0 00 0 00 0 00 Cli j 0 00 0 00 0 00 0 00 0 00 0 00 0 00 4 0 00 0 00 0 00 0 00 0 00 0 00 0 00 0 02 0 02 0 02 0 02 0 02 0 02 0 02 0 04 0 04 0 04 0 04 0 04 0 04 0 04 The Blurring Matrix Gli j is a N x N matrix where N 6 the matrix blurs the image and is known as a very effective one dealing with noises 0 04 0 04 0 04 0 04 0 04 0 04 0 04 0 04 0 04 0 04 0 04 0 04 Bii j 0 04 0 04 0 04 0 04 0 04 0 04 5 0 04 0 04 0 04 0 04 0 04 0 04 0 04 0 04 0 04 0 04 0 04 0 04 3 2 Reducing the Search Space Motion planning and autonomous decision models are often to be done in real time For that the algorithm must consider computation time constraints and optimize the CPU time at each time step Therefore the initial step is to recognize the interesting parts in the frame in our case the part that contains sea texture We characterize the sea texture by defin ing the horizon line and the horizon line orientation 9 The identification is com puted using the Canny Edge D
69. 0 8 1 0 SOG m s SOG m s speed over ground red speed over ground blue Q je oO Q Ww O N N Q 8 3 a LO T gt o 3 oO gt oo oO 2c Po us 3 D Q LO oO oO Ge er _ lt _ Damme 0 0 0 2 0 4 0 6 0 8 1 0 0 0 0 2 0 4 0 6 0 8 1 0 SOG m s SOG m s Fig 9 Histograms showing the typical boat speed distribution during the tests The plots indicate that the boats reach comparable speeds Note that the yellow boat started the course later and the wind calmed down towards the end of our test results Unfortunately we cannot expect the apparent wind angle to stay constant throughout the sampling windows e g due to residual errors in the boat control but also due to changes in wind direction and waves moving the mast and the wind vane To account for this we restricted the standard deviation for the apparent wind in the sampling window The left polar plot shows that upwind we can get fairly stable courses while the right plot indicates that it is particularly hard to maintain a constant apparent wind angle on a downwind course Furthermore the results show that the rrMM is capable of sailing at sustained speeds of almost 1 m s or 2kn As the boats are intended to have similar sailing performance we also summarize the boat speed for all four boats Figure 9 Clearly the histograms give only a rough idea of the sailing performance but given that the boats where certainly not perfectly trimmed the results indic
70. 1 3 Fuzzy Logic Controlled Sailing Boat by Abril et al The first documented results of fully autonomous sailing have been published by Abril et al 1 They presented a fuzzy logic controller for the rudder and sail of a sailing boat Test runs have been carried out on a yacht model with an overall length of 1 03 m a displacement of 4 5 kg and a sail area of 36 6 dm 3 1 4 Atlantis The Atlantis project of Stanford University began in 1997 with the concept of an unmanned autonomous GPS guided wing sail propelled sailing boat The boat is based on a Prindle 19 Catamaran with a self trimming wing sail FigureAeP The maiden voyage took place in Redwood City Harbour in January 2001 21 221 8 Gabriel Elkaim www A TIBOOK ir History and Recent Developments in Robotic Sailing 11 3 2 Competitions in Robotic Sailing In many fields of robotics competitions with memorable goals cause the attention of media and the interested public and therefore give a strong incentive to research and development in the particular area The most popular examples in robotics are DARPA Grand Challenge for completely autonomous cars 50 or RoboCup soc cer robots which aim to beat the human world champions by 2050 38 The same happened to autonomous sailing during the first decade of the 21 century when different events have been founded almost at the same time 3 2 1 Microtransat Research in autonomous sailing has been recently stimulated b
71. 11b wireless network card emulated a wireless access point and on MOOPI which used a newer Gumstix Verdex and had no compact flash option an ethernet cable connected the Gumstix to a DLink DWL G730AP access point In previous work it had been found that placing an access point on the robot provided a more robust solution than having the robot act as a client to a shore based access point as doing so meant that the human operator of a laptop was responsible for re establishing connection rather than relying on this process being automated by the robot A single UDP broadcast telemetry packet was sent at www A TIBOOK ir 44 C Sauz and M Neal a rate of 1 Hz This packet contained the current actuator positions GPS location the location of and heading distance to the next waypoint compass heading wind direction and if the course was directly sailable or not This data was also logged to a file upon every iteration of the main loop for replaying later on The data could be received by a remote laptop and displayed upon a moving map application 11 The use of a UDP broadcast packet allowed telemetry to be received despite high packet loss rates for multiple users to receive the data and for data to be received by systems which do not even have an IP address on the boat s subnet as obtaining a DHCP lease from the boat at long distance can be problematic The use of a Gumstix and wireless network came at quite a high cost to power consumptio
72. 2nd edn Academic Press New York 1997 10 Marham K C Clutter metrics for target detection systems IEEE Trans Aerosp Elec tron Syst 30 81 91 1994 11 Vijaya K A tutorial survey of composite filter designs for optical correlators Appl Opt 31 4773 4798 1992 12 Flannery D Loomis J Milkovich M Transform ratio ternary phase amplitude filter formation for improved correlation discrimination Appl Opt 27 4079 4083 1988 13 http sites google com site orenusv home research 1 usvvision www A TIBOOK ir Rule Compliant Navigation with Qualitative Spatial Reasoning Diedrich Wolter Frank Dylla and Arne Kreutzmann Abstract We develop a formal symbolic representation of right of way rules for sea navigation based on a qualitative spatial representation Navigation rules spec ified qualitatively allow an autonomous agent consistently to combine all rules applicable in a context The focus of this paper is to show how the abstract rule spec ification can be used during path planning We propose a randomized qualitative approach to navigation integrating the symbolic level with a probabilistic roadmap planner The resulting navigation system maneuvers under the side constraint of rule compliancy Evaluating our approach with case studies we demonstrate that quali tative navigation rules contributes to autonomous sailing 1 Introduction A considerable amount of everyday behavior is not self determined but
73. 484 usna edu phmiller usna edu www A TIBOOK ir 184 P Gibbons Neff and P Miller possibility of success Recognizing that average weather predictions are often quite different from short term forecasting a commercial route optimization program was used with real time data as a test of the route planning that would occur immedi ately prior to departure Route planning is used by every competent navigator prior to departure and now is commonly updated with the latest weather reports through out the voyage Long term planning that which occurs more than a few months prior to the voyage relies primarily on average climatological data Today s mo tor vessels are mostly concerned with avoiding bad weather and taking advantage of favorable currents to reduce fuel consumption Sailing vessels additionally must take advantage of favorable winds Cornell s World Cruising Routes 1 is a typical reference that provides traditional sailing routes developed over centuries of experi ence These traditional routes and their seasonal timing take advantage of prevailing winds currents absence of ice and minimal probabilities of storms The start of this project reviewed the traditional routes across the North Atlantic for small sailing craft The safest routes were Northern Route July August Newfoundland to Ireland or Portugal Southern Route November December Southern Europe to the Caribbean The question became which had the higher probabi
74. 5 doi 10 105 1 j3ea 200603 1 4 Stelzer R Jafarmadar K Communication architecture for autonomous sailboats In International Robotic Sailing Conference IRSC Porto Portugal July 2009 5 Sliwka J Reilhac P Leloup R Crepier P Malet H D Sittaramane P Bars F L Roncin K Aizier B Jaulin L Autonomous robotic boat of ensieta In 2nd Interna tional Robotic Sailing Conference Matosinhos Portugal 2009 6 Thomas G A Harris D d Armancourt Y Larkins I The Performance And Con trollability Of Yachts Sailing Downwind In Waves In Proceedings of the 2nd High Performance Yacht Design Conference Auckland New Zealand p 8 Ferburary 2006 7 Miller P Brooks O Hamlet M Development of the USNA SailBots ASV 2009 8 Roncin K Kobus J M Iachkine P Barr S M thodologie pour la validation du simulateur de voilier par des essais en mer une premi re tentative In Workshop Science Voile Ecole Navale LanveocPoulmic France 2005 9 Clarke D Gedling P Hine G The application of manoeuvring criteria in hull design using linear theory Trans RINA 125 45 68 1983 10 Van Oossanen P Theoretical estimation of the influence of some main design factors on the performance of international twelve meter class yachts In Presented at The Chesa peake sailing yacht Symposium Annapolis Maryland January 20 1979 www A TIBOOK ir www ATIBOOR ir A New Class for Robot
75. Dv2 for Source Specific Multicast 2006 6 Larson J Bruch M Ebken J Autonomous navigation and obstacle avoidance for unmanned surface vehicles In SPIE Unmanned Systems Technology VIII Orlando FL 2006 7 Navico Broadband Radar TM The Essential Guide 2009 8 Russel Technologies Inc RTI Radar Installation Maintenance Manual XIR3000C 2009 http www russelltechnologies ca downloads 9 Sauze C Neal M A raycast appoach to collision avoidance in sailing robots In In ternational Robotic Sailing Conference 2010 Proceedings pp 26 33 2010 10 Stelzer R Jafarmadar K Hassler H Charwot R A reactive approach to obstacle avoidance in autonomous sailing In International Robotic Sailing Conference 2010 Pro ceedings pp 34 40 2010 www A TIBOOK ir www ATIBOOR ir Route Planning for a Micro transat Voyage Peter Gibbons Neff and Paul Miller Abstract Many design decisions for an autonomous crossing of the North Atlantic Ocean require an understanding of both the time that the vessel will need to be self sufficient and the environmental conditions expected Potential trans Atlantic routes were evaluated to determine both the design conditions and selecting a route that has a relatively high probability of success Factors included prevailing winds currents ice gales calms sea state sunlight starting date boat characteristics and ship traf fic Two methods were used an evaluation of numerou
76. EE International Conference on Robotics and Automation ICRA pp 2372 2377 2006 www A TIBOOK ir Rule Compliant Navigation with Qualitative Spatial Reasoning 155 3 Bhatt M Loke S Modelling dynamic spatial systems in the situation calculus Spatial Cognition and Computation 8 1 86 130 2008 4 Cohn A G Renz J Qualitative spatial representation and reasoning In van Harmelen F Lifschitz V Porter B eds Handbook of Knowledge Representation pp 551 596 Elsevier Amsterdam 2007 5 Dylla F Qualitative spatial reasoning for navigating agents behavior formalization with qualitative representations In Gottfried B Aghajan H K eds BMI Book Am bient Intelligence and Smart Environments vol 3 pp 98 128 IOS Press Amsterdam 2009 6 Dylla F Frommberger L Wallgr n J O Wolter D Nebel B W lfl S Sailaway Formalizing navigation rules In Proocedings of the AISB 2007 workshop on Spatial Reasoning and Communication pp 470 474 2007 7 Freksa C Conceptual neighborhood and its role in temporal and spatial reasoning In Singh M G Trav amp Massuyes L eds Proceedings of the IMACS Workshop on Deci sion Support Systems and Qualitative Reasoning pp 181 187 Elsevier North Holland Amsterdam Amsterdam 1991 8 Kavraki L Svestka P Latombe J C Overmars M Probabilistic roadmaps for path planning in high dimensional configuration spaces IEEE Transactions on Robotics
77. Germany e mail firstname lastname uni ulm de firstname lastname uni ulm de Roland Stelzer INNOC Austrian Society for Innovative Computer Sciences Vienna Austria e mail Eirstname lastname innoc at www A TIBOOK ir 196 J Langbein R Stelzer and T Fr hwirth durations such as regattas over a few miles 12 In contrast for long term missions like ocean crossings weather conditions cannot be assumed to remain stable until the boat reaches its target Therefore a global view with consideration of weather forecasts is necessary In this paper we introduce a long term weather routing algorithm for autonomous sailboats and show how rule based programming facilitates a declarative and effi cient implementation In Section 2 we present how we modeled the sailboat routing problem and introduce our routing algorithm Its implementation is then discussed in Section B In Section 3 1 we compare our algorithm to existing commercial so lutions and discuss related work We conclude in Section 4 which also gives an outlook on future work 2 The Routing Algorithm Routing for sailboats no matter whether they are autonomous or not can be defined as the procedure where an optimum track is determined for a particular vessel on a particular run based on expected weather sea state and ocean currents 11 In this section we will take a closer look at the parameters required to find an optimum track and present our routing algo
78. MBit gross speed by the manufacturer and can be overwritten only in some products The achievable net speed under good conditions is often approximated by dividing the gross speed by half In either case the image data will use almost all airtime or even more Lost data is inevitable under such conditions Raising the mul ticast speed where possible on the other hand will prevent clients from associating with that access point in greater distance or bad radio conditions As aconsequence for the ASV Roboat we either need an access point that filters multicast or we need to split up the network into an internal and an external one The latter to be accessible via WiFi for monitoring purposes 6 2 Register Control Multicast address 236 6 7 10 port 6680 is used to set and read control registers in the radar Return messages seem to be sent via 236 6 7 9 port 6679 but as they are not needed for the basic operation we have not investigated the return channel so far An example can be found in table 2 Some operations require to set multiple registers in which case multiple UDP datagrams are sent The names of the registers below are arbitrary We tried to match them with their apparent function 6 2 1 Register 01 02 Radar Operation Payload length 1 byte To turn the radar on register 0x00 and 0x01 need to be set to 1 This starts the image data stream Likewise to turn the radar off both registers have to be set to 0 Sometimes the first
79. OOK ir History and Recent Developments in Robotic Sailing 23 53 Velagic J Vukic Z Omerdic E Adaptive fuzzy ship autopilot for track keeping Control engineering practice 11 4 433 443 2003 http 202 114 89 60 resource pdf 2235 pdf 54 Worsley P Self Trimming Self Tending Wingsails 2011 http www ayrs org wingsails pdf accessed on May 8 2011 55 Yeh E Bin J Fuzzy control for self steering of a sailboat In Proceedings of Singa pore International Conference on Intelligent Control and Instrumentation SICICI 1992 vol 2 pp 1339 1344 IEEE Los Alamitos 1992 http ieeexplore ieee org xpl freeabs_all jsp arnumber 637738 56 Zirilli A Tiano A Roberts G Sutton R Fuzzy course keeping autopilot for ships In 15th Triennial World Congress Barcelona Spain 2002 http www nt ntnu no users skoge prost proceedings ifac2002data content 00824 824 pdt www A TIBOOK ir www ATIBOOR ir Part II Robotic Sailboats www ATIBOOR ir www ATIBOOR ir Sailing without Wind Sensor and Other Hardware and Software Innovations Jan Sliwka Jeremy Nicola Remi Coquelin Francois Becket de Mesille Benoit Clement and Luc Jaulin Abstract In our paper we propose a solution for a robotic sailboat which is adapted to long travel across the ocean We tried to be as innovative as possible and stray out of the classical designs In fact robots often have different shapes from vehicles for mank
80. Operations Research 38 3 426 438 1990 7 Philpott A Mason A Optimising Yacht Routes under Uncertainty In Proceedings of the 15th Chesapeake Sailing Yacht Symposium CSYS 2000 2000 8 SailFast LLC SailFast Version 5 1 2011 9 Sailport AB Sailplanner 2011 10 Sneyers J Schrijvers T Demoen B Dijkstra s Algorithm with Fibonacci Heaps An Executable Description in CHR In Proceedings of the 20th Workshop on Logic Pro gramming WLP 2006 2006 11 Spaans J A Windship routeing Journal of Wind Engineering and Industrial Aerody namics 19 215 250 1985 12 Stelzer R Pr ll T Autonomous Sailboat Navigation for Short Course Racing Robotics and Autonomous Systems 56 7 604 614 2008 www ATIBOOR ir 204 J Langbein R Stelzer and T Fr hwirth 13 Stentz A Optimal and efficient path planning for unknown and dynamic environments International Journal of Robotics and Automation 10 3 89 100 1993 14 A Guide to the Code Form FM 92 IX Ext GRIB Edition 1 1994 http www wmo int pages prog www WMOCodes Guides GRIB GRIB1 Contents htm1 accessed November 28 2010 www A TIBOOK ir Author Index Ammann Nikolaus 157 Becket de Mesille Francois 27 Beckmann Daniel 71 Bishop Bradley E 87 Bouvart Gabriel 55 Bradshaw Joseph 87 Bruder Ralf 71 157 Busch Sebastian 169 Clement Benoit 27 Coquelin Remi 27 Dabrowski Adrian 169 De Malet Henry 55 Douale Nicolas 55 Dylla
81. Pradalier C Siegwart R Y Design and construction of the autonomous sailing vessel avalon In 2nd International Robotic Sailing Conference pp 17 22 2009 GRAUPNER GmbH amp Co KG K G http www graupner de fileadmin downloadcenter 2014 C_racing_MM Carbon_Edition_de_en_fr pdf cited May 31 2011 Klinck H Stelzer R Jafarmadar K Mellinger D Aas endurance An autonomous acoustic sailboat for marine mammal research In 2nd International Robotic Sailing Conference Matosinhos Portugal pp 43 48 2009 Miller P Beal B Capron C Gawboy R Mallory P Ness C Petrosik R Pryne C Murphy T Spears H Increasing performance and added capabilities of usna sail powered autonomous surface vessels asv In 3rd International Robotic Sailing Confer ence Kingston Ontario Canada pp 57 62 2010 Miller P Brooks O Hamlet M Development of the usna sailbots asv In 2nd Inter national Robotic Sailing Conference Matosinhos Portugal pp 9 16 2009 rrMM webpage http www r2m2 orgl cited May 31 2011 Sauze C Neal M Design considerations for sailing robots performing long term au tonomous oceanography In International Robotic Sailing Conference Breitenbaum Austria pp 21 29 2008 Sauze C Neal M Moop A miniature sailing robot platform In Proceedings of the 4th International Robotic Sailing Conference 2011 Stelzer R Pr ll T John R Fuzzy logic contro
82. S and one in open water to compare the running time of the two commercial solutions to our algorithm The tests were carried out on a 2 0 GHz Intel Dual Core with 4 GB of RAM while no other applications where running and the wall clock time taken for the computation was measured The output of Sailplanner indicates a grid width of about 30 km and 20 km for the resolutions Medium High and Ultra High respectively Hence we chose 30 km and 20 km as the grid resolution for the runs of our algorithm SailFast is fixed to a 6 hour resolution for the isochron lines in the demo version The results of the comparison are given in Table The results show that our algorithm is considerably faster when calculating routes A reason for this the could be the use of a heuristic function in our algorithm or the fact that SailFast and Sailplanner seem to consider points on land in the rout ing as well while our algorithm avoids them However the most likely reason is the fact that more than eight different bearings for each point are considered by Sail Fast and Sailplanner in contrary to our algorithm The results in Table I also indicate the expected trade off between computation time and quality of the approximation to the optimal route stemming from the complexity of the A algorithm which is exponential in the number of way points in the solution and thus the resolution of the grid 5 The routes computed by our algorithm and SailFast are almo
83. S simplified rules 11 18 These rules describe how vessels have to behave if they are in sight of each other in or der to avoid collisions The rules in the COLREGS are given in natural language using abstract spatial terms Additionally textbooks show pictorial representations in order to give people a more vivid interpretation about conditions and execution of rules For deriving rule formalizations we follow the approach taken by Dylla et al 5 6 which employs OPR A and its conceptual neighborhood structure In extension we give here additional qualitative representations and consider sailing vessels as well Let us consider rule 12 a and its pictorial representations Fig 5 in head on and crossing situation Rule 12 a when two sailing vessels are approaching one another so as to involve risk of col lision one of them shall keep out of the way of the other as follows 1 when each of them has the wind on a different side the vessel which has the wind on the port side shall keep out of the way of the other ii In order to define rule consistent or rule compliant behavior we need to oper ationalize the individual rules First we translate and ground the natural language terms in qualitative relations and second formalize the rules by means of this repre sentation Finally we will define rule compliant behavior based on these formaliza tions By using OPR An relations we abstract from the physically extended o
84. SKAMP The first attempt to autonomous sailing recorded in the literature is a project named SKAMP Station Keeping Autonomous Mobile Platform The SKAMP was a wind propelled mobile surveillance platform and utilized a curving ring shaped rigid wing sail Figure 4 a It was developed in 1968 by E W Schieben with the Radio Cor poration of America and was optimised for autonomous station keeping rather than for dynamic performance 145 Actual sailing data have never been published so it remains unclear whether SKAMP ever sailed autonomously 21 http www microtransat org www A TIBOOK ir 10 R Stelzer and K Jafarmadar b Fig 4 Early robotic sailing boats a SKAMP Station Keeping Autonomous Mobile Plat form b RelationShip c Atlantis 3 1 2 RelationShip The second published autonomous sailing attempt was the RelationShip project of University of Applied Science in Furtwangen Germany Figure 4b The project started in 1995 with the ambitious plan to sail around the world with an unmanned trimaran According to Elkaim the initial intention was to sail autonomously However after some difficulties the project changed to a remote control via satel lite After some years the project was cancelled due to regulatory difficulties They did not get the permission to circumnavigate the globe with their unmanned Rela tionShip The idea to declare the boat as floatsam did not convince the maritime authorities 47 3
85. Spengergasse e mail dabrowski spengergasse at Sebastian Busch Adrian Dabrowski Austrian Society for Innovative Computer Sciences e mail imail sebastianbusch at adrian innoc at Roland Stelzer Head of Austrian Society for Innovative Computer Sciences e mail roland innoc at www A TIBOOK ir 170 A Dabrowski S Busch and R Stelzer ir eh De er Fig 1 A photograph of the display unit and a screenshot of the JAVA program showing the same data surroundings with a birds eye view accurate range measurement and navigability under poor visual conditions 2 Motivation The ASV Roboat 2 needs a radar solution which provides the control algorithms with dynamic information about the surrounding real world in order to implement obstacle avoidance and possibly to handle other navigational and mapping tasks The solution should preferably deliver its data directly in digital form The developers of project ROAZ nu as well as INNOC independently con tacted several vendors of small radar systems but had no success in getting interface specifications even when offered to sign an NDAR Some vendors have their own PC Windows software on sale others presumably want to keep that option for the future Russel Technologies Inc former Xenex Innovations Ltd sells a commercial PC Windows based solution 8 including an SDK for Furuno Radars The System consists of an embedded PC which digitizes the proprietary Furuno video signal as
86. The means by which students carry out their project are an end of sorts of their own providing students with experience and insight into development of a com plex system It is unlikely that any student will be building robot sailboats for a living at any point after graduation so the main outcome of the effort must of needs be a deeper understanding of the engineering process in general It is true however that robotic sailing is a highly specialized task that involves all of the disparate aspects of systems engineering from power through expert control As students progress through the tasks associated with this project there are several key points at which domain specific lessons can be driven home tying the design experience more closely to the students core discipline and coursework Functional Blocks Because the task of generating a complete sailing robotic sys tem in just one semester is quite daunting the planning phase becomes of vital importance Students cannot afford the luxury of working together on every aspect of the task and so must divide the work Because of the tightly integrated nature of the systems each sub team must understand and work to a specific set of interface and power requirements As such every subsystem is managed as a functional block in the overall system Each sub team leader is responsible for maintaining an accurate list of input requirements output format power require ments and total volume
87. a dummy terminal using HyperTerminal as the attitude of the sensor was adjusted on the gimbal The results of the test showed that the system could relia bly measure airflow with a relative angle of up to 30 degrees The test did NOT however show that the measurements were accurate A good bench test would have a means of verifying the measured data 2 Interoperability test Here each functional component or subsystem that in teracts with any additional component s is tested with each This is done first with each communicating component individually and then by adding compo nents subsystems individually as is reasonable and achievable extremely complex systems have a combinatorial explosion in this test and will require additional care in selecting subsystems to test When and if failures occur in dividual components can be isolated for further analysis and troubleshooting This set of tests is only complete when the entire fully connected system is functional with stable wall power It is crucial to point out that the system does not need to be fully programmed to test the interoperability but that any and all communication modalities and control signals must be exercised to show their operation It is typical to start with the processor in this test adding pe ripheral components one at a time 3 Powered tests Using the designed power supply each component is again tested to show full performance is met This step is optional an
88. a few minutes and establishing the direction of travel It was discovered during this experiment that it was possible to steer the boat onto any point of sail but that it was difficult to remain on course when sailing downwind Typically after only a few seconds of downwind sailing the boat would turn towards the wind It was also found that the boat would not always settle on a predictable course for a given set of sail positions and that there were often multiple possible courses including sailing backwards that could be sailed for any given position A video of this experiment is available from 2 MOOPn was developed with the intention of using a rudder and twin wing config uration together Its control system was implemented to use the sails to assist in the steering process Only a total of 11 unique positions were allowed for each sail this design had been inherited from MOOPO and MOOPI which suffered considerably from poor repeatability of their sail servos and 11 positions had been the maximum that could be achieved in a consistently repeatable manner The sail assisting pro cess used the rear sail only and would move it by a maximum of one position If the boat was heading too close to the wind then the sail would be let out one position and if it was too far from the wind the sail would be pulled in by one position The sails would also be placed into a goose swing position one sail on each side when the wind was coming from a 20
89. a grid of wind vectors w containing the wind speed in north and east direction A GRIB file can contain multiple forecasts which are made available for up to 16 days in intervals as small as three hours The resolution of the wind data typically ranges between 0 5 and 2 5 degrees 2 1 2 Sailboat Behavior To calculate the time required to travel between two waypoints we need to know the speed of the sailboat for given wind conditions This speed can be described as a function of the wind speed and the angle between the wind and the boats heading that is to say the boats velocity v v w if denotes the true wind angle of the boat This function is usually shown in a plot known as polar diagram Figure l shows the normalized polar diagram of the ASV Roboat 12 which describes the relation between wind speed and boat speed for a given true wind angle Another factor to take into consideration is the so called hull speed which we treat as the approximative maximum speed of the boat In our algorithm this maximum speed is configurable by the user We approximate the travel time 1 between two locations x and x on a great circle path by taking the wind conditions w and true wind angle at x for the first half of the leg and the wind conditions w and true wind angle a at x for the sec ond half The distance d between x and x is calculated using the laws of spherical geometry 2 Together we get the travel time t 5 dij
90. a sailing robot without a wind sensor It was found that the Futaba S3306 servos used in the previous MOOPs had only about a 200 range of turn and that this was insufficient for the sail positions required to sail without a rudder To overcome this the servos were modified to allow 360 rotation a continuous rotation potentiometer was placed around the shaft and a Polulu Qik 2s12v10 motor controller was used to control the motor from the PIC In summer 2010 MOOP3 was entered into the World Robotic Sailing Championships and Sailbot competition held in Kingston Ontario Canada Before entering it underwent a few changes a Gumstix was added to allow it to run the same control system software as MOOPO and MOOP I and a wind sensor was added As there was no wind sensor cabling in the masts sails the wind sensor was placed on a pole extending up from the deck hatch A photograph of MOOP3 can be seen in Figure and its full specifications are shown in Table 3 2 4 MOOPn MOOPn was constructed in Spring 2010 and sold to Nottingham University It fea tured twin wing sails a rudder of the same design as MOOPI and MOOP2 its wind sensor used an AS5040 magnetic encoder as used by MOOPO and MOOP1 a Gum stix single board computer and DLink pocket access point were used The internal layout of the boat was slightly different to all previous boats All electronics except the rudder servo were attached to a plastic shelf inside the hull which in turn was
91. acommon Ethernet switch www A TIBOOK ir 1 2 A Dabrowski S Busch and R Stelzer which we placed in between Radar Interface Box and the Display and Control Unit We connected a notebook to the hub and used Wireshark to record and analyse the network trafi dl The same setup has been used to test our implementation of the display and con trol functions as we could simultaneously use the proprietary controller and display and our application to compare the results To test which data streams are essential and which not a computer with two Ethernet interfaces and a software bridging solution has been used It allowed us to filter out specific traffic based on layer 3 and 4 attributes without breaking the broadcast domain 5 Network Communication During boot up the radar antenna transceiver and the display unit automatically con figure their IP addresses If no DHCP server is available both devices select an IPv4 Zeroconf link local address as described in RFC3927 4 All communication is done using UDP multicast datagrams All units emit Internet Group Management Proto col IGMPv3 5 subscription announcements This approach has two advantages 1 It enables the system to handle multiple subscribers 1 e controller display units and 2 The IP addresses of the devices are irrelevant since they solely communicate through fixed multicast groups 6 Control and Data Streams Figure 3 shows the 11 data streams we identified on both
92. aged pixels will always be a part of the image We assume that the basic structure of the target does not change from one frame to another The algorithm characterizes a target by using a Skeleton algorithm This algorithm applies a thinning process to an image and finds a simpler structure of the recognized targets This process simplifies the target tracking in the next frames and decreases the CPU time We use an extended version of the Skeleton algorithm The extended version is based on doubled thinning process of the target For each pixel x belongs to the thin target profile the pixel at x 1 location also marked as a pixel in a thin profile of the target The left side of Figure 5 shows the target image before the thinning process and the right side shows the thin target after applying the Skeleton algorithm Fig 5 Skeleton Algorithm The left side shows the target image before the thinning process and the right side shows the thin target after Skeleton algorithm without Interesting Point www A TIBOOK ir Automatic Obstacle Detection for USV s Navigation Using Vision Sensors 135 Marine environment can be unstable and thin structure targets usually can t be seen in the frame Therefore the target structure should be more flexible supporting target recognition even if not all of the target is in the frame For that we define the Interesting Point concept Definition Interesting Point is a point in the target thin profile wh
93. ailing Roland Stelzer and Kar m Jafarmadar Abstract Robotic sailing boats represent a rapidly emerging technology for various tasks on lakes and oceans In this paper we give an overview about the main build ing blocks of a robotic sailing boat for controlling the rudder and the sails History of robotic sailing includes developments in mechanical electronic and intelligent self steering systems as well as automatic sail control Furthermore advantages and disadvantages of rigid wing sails in comparison to traditional fabric sails are illumi nated Early examples of robotic sailing boats and recent developments stimulated by robotic sailing competitions such as Microtransat Challenge SailBot and World Robotic Sailing Championships are presented We conclude with a brief outlook on potential applications in the field of robotic sailing 1 Introduction Autonomous sailing robots perform the complex tasks of sailing boat navigation fully automatically and without human assistance Bowditch defines navigation as the process of monitoring and controlling the movement of a craft or vehicle from one place to another Robotic sailing boats therefore have to perform the complex planning and ma noeuvres of sailing fully automatically and without human assistance Starting off Roland Stelzer INNOC Austrian Society for Innovative Computer Sciences Haussteinstra e 4 2 1020 Vienna Austria e mail roland stelzer innoc at Centre for Co
94. ailing yacht In Proceedings of the International HISWA Symposium on Yacht Design and Yacht Construction 2004 Roncin K Kobus J M Dynamic simulation of two sailing boats in match racing Sports Engineering 7 3 139 152 2004 Sauz C Control software for a sailing robot Master s thesis University of Wales Aberystwyth 2005 Sauz C Neal M A neuro endocrine inspired approach to long term energy autonomy in sailing robots In Proceedings of TAROS Towards Autonomous Robotic Systems 2010 Sauz C Neal M Long term power management in sailing robots In Proceedings of IEEE Oceans 2011 2011 www A TIBOOK ir Part IV Collision Avoidance www A TIBOOK ir www ATIBOOR ir Automatic Obstacle Detection for USV s Navigation Using Vision Sensors Oren Gal Abstract This paper presents an automatic method to acquire dentify and track obstacles from an Unmanned Surface Vehicle USV location in marine environ ments using 2D Commercial Of The Shelf COTS video sensors and analyzing video streams as input The guiding line of this research is to develop real time au tomatic identification and tracking abilities in marine environment with COTS sen sors The output of this algorithm provides obstacle s location in x y coordinates The ability to recognize and identify obstacles becomes more essential for USV s autonomous capabilities such as obstacle avoidance decision modules and other Artificial
95. algorithms rather than on boat design So far two commercially available hulls have been adapted for the purpose of autonomous sailing Roboat I Figure 6 is based on a ready made yacht model of type Robbe Atlantis intended to be remote controlled It is a gaff rigged schooner 3 with alength of 1 38 m a height of 1 73 m and a total displacement of 17 5 kg The four sails comprise a total area of 85 5 dm The boat is equiped with a 800 MHz PC running Linux a GPS receiver a tilt compensated electronic compass and sensors for wind speed and wind direction Roboat I won the Microtransat competition in 2006 The basis for the ASV Roboat Figure Bil is the commercially available boat tyle Laerlind The boat was originally created for kids to learn sailing and there fore safety and stability are its major characteristics It has a length of 3 75 m and comprises a 60 kg keel ballast which will bring the boat upright even from the most d Fig 7 Autonomous sailing vessels with a length of exactly 2 m a Black Adder Queen s University b First Time USNA c Gill the Boat USNA d Luce Canon USNA 12 INNOC 13 A schooner is a type of sailing vessel characterized by the use of two or more masts with the forward mast being no taller than the rear masts 14 INNOC 15 http www laerling nl www ATIBOOR ir History and Recent Developments in Robotic Sailing 15 b d Fig 8 Autonomous sailing vessels with
96. alse Sperry E Automatic steering Society of Naval Architects and Marine Engineers 1922 Spiegel Fliegender Schwarzw lder Der Spiegel 22 176 177 1998 http wissen spiegel de wissen image show html did 7894363 amp aref image017 SP1998 022 SP199802201760177 pdf amp thumb false Stelzer R Jafarmadar K Communication architecture for autonomous sailboats In Proceedings of International Robotic Sailing Conference Matosinhos Portugal pp 31 36 2009 ftp www inc eng kmutt ac th pornpoj SailBoat 1266686522 pdf Stelzer R Proll T John R Fuzzy logic control system for autonomous sailboats In FUZZ IEEE 2007 London UK pp 97 102 2007 http dx doi org 10 1109 FUZZY 2007 4295347 Thrun S Montemerlo M Dahlkamp H Stavens D Aron A Diebel J Fong P Gale J Halpenny M Hoffmann G et al Stanley The robot that won the DARPA Grand Challenge In The 2005 DARPA Grand Challenge pp 1 43 2007 http citeseerx ist psu edu viewdoc download doi 10 1 1 111 1920 amp rep repl amp type pdf Van Aartrijk M Tagliola C Adriaans P AI on the Ocean the RoboSail Project In ECAI pp 653 657 Citeseer 2002 http citeseerx ist psu edu viewdoc download do1 10 1 1 84 8172 amp rep repl amp type pdf Van Amerongen J Adaptive steering of ships A model reference approach Automatica 20 1 3 14 1984 http www ce utwente nl rtweb publications 1984 pdf files Automatica84 pdf www A TIB
97. an Autonomous Sailing Vessel 0 cece eee een Bradley E Bishop Joseph Bradshaw Cody Keef Nicholas Taschner Using ARM7 and uC OS II to Control an Autonomous Sailboat Michael Koch Wilhelm Petersen 71 www ATIBOOR ir X Contents Simulating Sailing Robots 0 00 00 Colin Sauze Mark Neal Part IV Collision Avoidance Automatic Obstacle Detection for USV s Navigation Using Vision SENSOT Se een ee Oren Gal Rule Compliant Navigation with Qualitative Spatial Reasoning Diedrich Wolter Frank Dylla Arne Kreutzmann Global Data Storage for Collision Avoidance in Robotic Sailboat Racing The World Server Approach 005 Nikolaus Ammann Florian Hartmann Philipp Jauer Julia Kr ger Tobias Meyer Ralf Bruder Alexander Schlaefer Part V Localization and Route Planning A Digital Interface for Imagery and Control of a Navico Lowrance Broadband Radar oo come Adrian Dabrowski Sebastian Busch Roland Stelzer Route Planning for a Micro transat Voyage 0 Peter Gibbons Neff Paul Miller A Rule Based Approach to Long Term Routing for Autonomous Sa DOA Sea g 2er hace ara ae Hake ae eee eee eae oe Johannes Langbein Roland Stelzer Thom Friihwirth Author Index rererere eae Sg daran Sa a ee www A TIBOOK ir Part I Introduction www ATIBOOR ir www ATIBOOR ir History and Recent Developments in Robotic S
98. an ultrasonic wind sensor using a single 40 kHz transmitter and two receivers One receiver picked up the signal in the X plane and the other in the Y plane By ampli fying the received signal combining it with the original signal ina NAND gate and charging a capacitor this produces a voltage proportional to the time of flight for each sensor Although this approach worked reliably under laboratory conditions it was exceptionally sensitive to temperature changes and water droplets on the sen sors It is discussed in more detail in 10 Eventually this sensor was replaced with a continuous rotation potentiometer and later a rotary magnetic encoder Work on MOOPI was begun in early 2009 and aimed to produce a simpler easier to construct hull A balsa wood plug carved using a similar principle of multiple cross sections that was used in MOOPO s polystyrene core Using balsa wood allowed a much smoother finish to be created than with the polystyrene The plug was then covered in glass fibre and epoxy to create a mould for half the hull from which other hulls could be made A photograph of this is shown in Figure 4 Once the two halves of the hull had been made they were glued together with epoxy resin The deck was cut from a single piece of acrylic plastic To prevent the deck flexing around the sail plastic blocks were placed around the mast between the sail servo and the deck A shelf was then created inside the hull where electronics could be placed
99. ance nm 130 Heading 12 True Wind Di force Rel WA VMG kt SOG kt Time hr rection N 5 4 12 3 5 3 4 38 4 NE 4 4 27 2 0 1 9 67 4 E 4 3 18 1 7 1 6 80 4 SE 7 4 63 3 4 32 40 1 S 26 4 108 3 3 3 2 41 1 SW 33 4 153 3 0 2 9 44 9 W 14 4 162 3 0 2 9 44 3 NW 5 4 117 3 3 32 41 1 Calm 2 min time 38 4 hours likely time 44 9 hours max time 80 4 hours current direction and velocity for each five degree segment Every segment was then broken up into the eight wind directions so that the time through the cell could be calculated The process for obtaining the time to pass through the cell for each wind direction began with calculating the relative wind angle Using north as a ref erence of zero degrees the wind direction was subtracted from the heading to give the relative wind angle Using the Beta angle from the appropriate VPP table for the wind Force gave the expected boat velocity The beginning and ending locations in the cell gave the distance through the cell To get the Speed of the Ground SOG the current vector for the cell was added to the boat speed using vector addition The distance through the cell divided by the predicted SOG gave the time through the cell for that wind direction In cases where the wind direction resulted in beating or running the appropriate Velocity Made Good was used with the condition that the boat would not sail closer to the wind than 46 degrees From the percentage like lihood of the wind direction
100. and Automation 12 4 566 580 1996 9 Kuipers B Qualitative Reasoning Modeling and simulation with incomplete knowl edge MIT Press Cambridge 1994 10 Moratz R Representing relative direction as binary relation of oriented points In Pro ceedings of the 17th European Conference on Artificial Intelligence ECAI 2006 Riva del Garda Italy 2006 11 Pommerening F W lfl S Westphal M Right of way rules as use case for integrating GOLOG and qualitative reasoning In Mertsching B Hund M Aziz Z eds KI 2009 LNCS vol 5803 pp 468 475 Springer Heidelberg 2009 12 Renz J Qualitative spatial and temporal reasoning Efficient algorithms for everyone In Proceedings of the 20th International Joint Conference on Artificial Intelligence IJCAI 2007 pp 526 531 2007 13 Renz J Nebel B Qualitative spatial reasoning using constraint calculi In Aiello M Pratt Hartmann I E van Benthem J F eds Handbook of Spatial Logics pp 161 215 Springer Heidelberg 2007 14 Smierzchalski R Michalewicz Z Modeling of ship trajectory in collision situations by an evolutionary algorithm IEEE Transactions on Evolutionary Computation 4 227 241 2000 15 Statheros T Howells G Maier K M Autonomous ship collision avoidance navigation concepts technologies and techniques Journal of Navigation 61 129 142 2008 16 Stelzer R Pr ll T John R I Fuzzy logic control system for autonomous sa
101. and arrival con ditions The actual selection also included an analysis of shipping traffic Both the Northern and Southern Routes have significant traffic although established shipping lanes and local traffic patterns are well documented and can be avoided Figure 5 for instance shows the shipping density around Newfoundland Choosing between the Northern and Southern Routes became a question of the probability of success and mitigating potential problems A first order approach used the algebra of ran dom variables based on the most likely voyage time For instance the reliability of the three prior USNA SailBots during sailing was calculated from testing logs as 0 9985 hr This would imply a probability of success for the Northern Route of 58 and for the Southern Route 33 The uncertainty however puts this in to question as the new boat uses different and supposedly more reliable equipment Addition ally while both routes have similar wind and sea conditions the Northern Route has some potential for ice and slightly higher traffic particularly near Ireland Using a linear probability for ice encounter based on the extent of the berg limit versus the pack ice line and the route through the ice zone the Northern Route drops to 49 Ice travels slowly however and short term forecasting can mitigate that problem Finally the Northern Route will provide only 50 probability of sunlight versus nearly 100 for the Southern Route 2 This may b
102. and strength the fastest slowest and most likely times through the cell were identified Table 2 shows a typical calculation for a cell 4 1 Possible Routes As mentioned above two main routes were identified The Northern Route was from Newfoundland to Ireland and the Southern Route was from Europe to the Caribbean www A TIBOOK ir 188 P Gibbons Neff and P Miller The main factors involved using the prevailing wind and current directions associ ated with the North Atlantic Ocean gyre and local coastal effects 4 1 1 Northern Route Originating in Newfoundland a northern route would have the shortest time at sea The island has numerous harbors that are suitable but the main harbor of St John s or the beach at Cape Spear has the greatest potential due to its logistics most easterly location and southerly breeze once past the headlands Pilot Chart analysis indicates a departure time during the last two weeks of July An earlier departure would in crease the risk of pack ice and leaving later would increase the possibility of both hurricanes and extended calms In both cases the dangerous area is the first 200 nautical miles Figure 2 shows a detail of the Pilot Charts for July and Figure 3 shows the probability of hurricanes in the North Atlantic 5 A more southern de parture location would eliminate ice exposure but would place the vessel in heavy trafficked shipping lanes Figure 4 illustrates the probable range of tracks and li
103. area We want to ensure that each participant has access to the same data Thus the chances of every participant in a fleet race are balanced Based on our experience with small boats we decided to use a centralized system Instead of each boat actively monitoring other objects we introduce the so called World Server This server gathers the sensor data sent by each boat and stores it Subsequently data from all vessels and other objects such as buoys coastline and obstacle areas are combined into one packet and provided upon request In this work we present our World Server approach This is motivated by an overview of two similar systems given in section 2 In section we discuss potential implementations and give reasons for our choice Subsequently our implementation of the World Server is described in section 4 To conclude our paper we present some results in section 5 2 State of the Art In this section we shortly present two existing system for vicinity awareness On the one hand there is the Automatic Identification System AIS which is used in commercial ship traffic On the other hand the RoboCup Soccer Server Simulator is described 2 1 Automatic Identification System The Automatic Identification System AIS is intended for tracking ships and used by Vessel Traffic Services VTS Important goals of VTSs are 5 MONITORING vessel movements INFORMING mariners of other vessels and potential hazards RECOMMENDING courses o
104. ary de sign Details of this process are beyond the scope of this work but can be found in 1 The advisor plays a crucial role in the development of the preliminary design Students will tend to gravitate toward well understood or proven components to fill out the morphological chart and will often resort to nonsensical entries to fill out a row For example when selecting components for the power system a typi cal morphological chart will show batteries unspecified and solar with the oc casional internal combustion or nuclear power option Students must be guided to look into battery technologies and study their characteristics to select an appropri ate suite of choices using the metrics and constraints Once a realistic set of options have been enumerated the students must rely on the available data and their engineering insight to select the preliminary design us ing a decision matrix Here the advisor must assist with realistic scores for met rics that may be difficult to predict Whenever possible direct experience with the systems under consideration is appropriate In 2011 the SailBot systems team de veloped constructed and tested two completely different and independent sensing systems in order to evaluate the efficacy of each Having selected a preliminary design for each subsystem a compatibility check is run If two subsystems are incompatible a decision must be made as to which to disallow Assuming that such a decisi
105. as modified to transmit the boat s position every few hours This was powered by a separate set of AA lithium batteries so that it would continue to function in the event of the rest of the control system failing The full specifications of MOOP2 are shown in Table 2 and a photograph of it sailing in the sea off Aberystwyth is shown in Figure 6 Fig 4 The plug left and the mold right for the MOOP hulls www A TIBOOK ir 46 C Sauz and M Neal Table 2 The specification of MOOP2 Name MOOP2 Date of Construction Summer 2009 Sails Single Wing Sail Sail Actuator Servo Rudder Actuator Servo on magnetic linkage Computers PIC18LF4550 Wind Sensor Furuno Rowind Batteries 10 13 Ah NiMH rechargeable size F batteries Compass HMC6343 GPS SiRF3 Globalsat EM 408 Other Solar Panels SPOT Notes Built to enter the Microtransat Challenge 2 3 Twin Wing Designs In summer 2009 work began on a twin wing sail MOOP known as MOOP3 Based upon work by Benatar Qadir Owen and Baxter 6 using the twin wing sails of a 1 5 m long sailing robot to perform steering instead of the using the rudder this boat was constructed with the idea of further investigating if the rudder could be removed By turning the rear sail into the wind and stalling it the boat would rotate away from the wind and by turning the front sail into the wind the boat would turn to wards the wind The primary motivation to removing the rudder was to eliminate the complexity
106. ate no fundamental difference in boat speed Note that not all boats where sailing at exactly the same time and the wind speed decreased towards the end of our tests The boats started in the order blue red green and yellow and the green boat had to abandon the test after approximately 20 min www A TIBOOK ir 82 Fig 10 Servos positions and sensor readings during a typical tack The top plot shows the angles of rudder main sail and jib over time with the vertical gray line indicating the approximate initiation of the tack Ini tially the boat is on a stable course and the main sail angle is approximately 15 larger than the jib angle During the tack the rudder turns the boat the main sail moves along with the apparent wind and the jib angle first increases and then maintains an offset until the boat heads straight into the wind This can be seen in the center plot where the blue line shows the apparent wind angle The compass readings also change im mediately while the GPS heading is subject to a sub stantial delay A similar delay can be seen in the boath speed in the bottom plot which also indicates that the boat did almost come to a complete stop The accelerometer readings indicate that the boat also heeled substantially which is likely exacerbated by backing the jib However the boat is on a stable course on the new tack after ap proximately 4s degrees 20 O0 degrees 50 20 40 60
107. ation NOAA Pilot Charts SOA s predicted performance in Force 4 is shown in Table 1 Predictions were made for windspeeds ranging from Force 1 through Force 8 Of importance in Force 2 which was a possibility for both the northern and southern routes SOA had a minimum predicted speed of 0 8 knots That is barely above the 0 6 knot maximum current seen on those routes The vessel would have a difficult time sailing against the current www A TIBOOK ir Route Planning for Micro transat Voyage 185 Table 1 Performance Prediction for Spirit of Annapolis in Force 4 Beta is the relative wind angle Vmdgd is the Velocity Made Good either directly upwind beating or downwind run ning beta deg Vel kt Heel deg Vmded kt 38 2 6 28 2 2 0 40 2 1 29 0 Pl 42 2 8 29 6 21 44 2 9 30 2 2 1 46 2 9 30 8 2 0 48 3 0 31 2 2 0 50 3 1 31 7 2 0 55 3 2 32 6 1 9 60 3 4 31 1 1 7 65 3 4 28 0 1 5 12 3 5 23 8 1 1 80 3 5 19 4 0 6 90 3 5 14 7 0 0 100 3 4 10 9 0 6 108 3 3 8 6 1 0 120 3 1 6 7 1 5 132 3 0 6 2 2 0 144 3 0 3 7 2 4 150 3 0 3 2 2 6 160 3 0 3 9 2 9 170 3 1 2 0 3 0 175 3 1 1 0 3 1 178 3 1 0 4 3 1 3 Planning Options Route planning took two time approaches The first was long range planning which relied on climatological data presented in NOAA Pilot Charts 2 The second was short term planning which relied on weather model predictions for 24 hours to two weeks and was available in GRIB format from the National Weather Service Both
108. ber 487979 25 Giger L Wismer S Boehl S Biisser G Erckens H Weber J Moser P Schwizer P Pradalier C Siegwart R Design and construction of the autonomous sailing vessel avalon In Proc 2nd Int Robotic Sailing Conf pp 17 22 2009 26 Layne J Passino K Fuzzy model reference learning control for cargo ship steer ing IEEE Control Systems Magazine 13 6 23 34 1993 http www2 ece ohio state edu passino PapersToPost FMRLC ship pdf 27 Loibner D KLAR zur WENDE klick Yacht Revue 3 1998 http www esys org stories trimaran html 28 Manley J Unmanned surface vehicles 15 years of development In OCEANS 2008 pp 1 4 IEEE Los Alamitos 2008 http www oceanicengineering org hi1story 080515 175 pd 29 Microtransat Official microtransat web site 2011 http www microtransat org accessed on 27 April 2011 30 Miller P Beal B Capron C Gawboy R Mallory P Ness C Petrosik R Increasing performance and added capabilities of usna sail powered autonomous surface vessels asv In International Robotic Sailing Conference 2010 http www dtic mil cgi bin GetTRDoc AD ADA534798 amp Location U2 amp doc GetTRDoc pdf 31 Miller P Brooks O Hamlet M Development of the usna sailbots asv In Inter national Robotic Sailing Conference 2009 http www dtic mil cgi bin GetTRDoc AD ADA534672 amp Location U2 amp doc GetTRDoc pdf 32 Minorsky N Directional s
109. bjects to oriented points Wether the extended objects collide or not will be handled in the planning phase www ATIBOOR ir Rule Compliant Navigation with Qualitative Spatial Reasoning 147 The start configurations for the application of rule 12 a is that two sailing ves sels are approaching one another so as to involve risk of collision For example this configuration is given if two vessels are meeting on reciprocal or nearly reciprocal courses which is described by relation 4x9 f Furthermore we need to know from which side of the vessel the wind comes from We represent the orientation of the wind by an OPRA same relation which is generated from the vessel s heading and the orientation of the wind as seen from the center point of the vessel In contrast to the representation of the relative orienta tion of the vessels we need the exact heading Therefore we cannot apply OPRAZ and must apply the original OPR A4 Thus if A 4Zi Aying with i 1 7 than the wind is coming from port and with i 9 15 its coming from starboard In summary the conditions of rule 12 a can be represented as A 4x9 BAA 4Zi Aying AB 4Zj Bying 1 with ie 1 7 and j 9 15 The evasion behavior needs to be modeled next Vessel A is the give way vessel and B the stand on vessel Regarding the advised behavior as shown in Fig B must keep its course and A must turn starboard in order to avoid a collision Considering a
110. cability of symbolic planning can be questioned www A TIBOOK ir Rule Compliant Navigation with Qualitative Spatial Reasoning 143 navigation situation qualitative navigation control gt sation gt roadmap planner output gt rules auaitatveh qualitative control u abstration abstration physical world amp physical simulation wind me gt on NS Fwindl f Fig 1 Tacking a sailboat Fig 2 Architecture overview By contrast we use the symbolic level to formalize navigation rules as sequences of key configurations to pass through avoiding action definitions Key configura tions are used as intermediate goals in a probabilistic roadmap planner Probabilistic planning has previously been shown to be applicable in traffic planning In contrast to the approach by 14 we explicitly model collision regulations in a formal lan guage The combination of qualitative representation with probabilistic roadmap planning has also been suggested in 17 but our approach does not need to gener ate a plan on an abstract level before invoking the action planner This way we avoid the aforementioned problem of exactly describing action preconditions Figure 2 presents an overview of our approach Based on the qualitative assessment of an ob servation we select the navigation rules applicable to the situation Then we employ the planner to determine control actions for rudder and sheet rope length that allow the vessel to navigat
111. can be used www ATIBOOR ir Rule Compliant Navigation with Qualitative Spatial Reasoning 153 Table 1 Analysis of plans obtained for the test scenarios scenario active nodes success rate path length m plan duration s 1 20 86 180 6 4405 64 6 431 1 1 100 90 152 2 16 0 57 5 455 0 1 200 93 138 8 439 8 78 0 472 0 2 20 22 201 5 21 9 70 0 416 5 2 100 48 223 3 23 0 68 0 14 8 2 200 63 214 8 26 4 62 1 14 8 3 20 81 106 3 19 0 38 5 8 0 3 100 84 105 5 49 2 34 5 46 5 3 200 84 102 9 45 6 31 6 45 5 4 20 38 94 2 12 4 109 8 74 1 4 100 66 95 8 13 9 58 2 433 0 4 200 79 97 13 16 0 60 2 39 8 to balance the need of restarts with the computational demands For example con sidering the simple scenario it can be seen that that a set of only 20 active nodes gives a success rate of 86 which by using ten times the amount of active nodes can only be increased to 93 Thus using a small set of active nodes and restart ing if necessary provides efficient means for path planning The measured standard deviation shows that the overall navigation quality shortest quickest route leaves room for improvement in particular the high standard deviation in the easy scenario results from some outliers in terms of long detours However the performance of our simple metric planning systems already indicates that a randomized qualitative approach enables rule compliant navigation In particu
112. competitions on the water and a scientific confer ence provides an opportunity to practically demonstrate theoretical developments 3 3 Competing Teams and Their Sailing Robots This section provides an overview of the teams which participated in recent robotic sailing competitions and are covered by scientific literature Many of them have been encouraged by these events to start research in the field of autonomous sailing Figure 5 shows the increasing number of boats competing in autonomous sailing competitions since their invention in 2006 The teams are presented here in alpha betical order H Microtransat E SailBot EIWRSC Number of boats 2006 2007 2008 2009 2010 Year Fig 5 Number of boats competing in Microtransat SailBot and WRSC In 2010 SailBot and WRSC were organised as a single event http www roboticsailing org www ATIBOOR ir History and Recent Developments in Robotic Sailing 13 b c d 2 Fig 6 Autonomous sailing vessels with a length of less than 2 m a D umling University of L beck b MOOP University of Aberystwyth c Pi mal Daumen University of L beck d Breizh Spirit ENSTA Bretagne e Roboat I INNOC f AROO University of Aberystwyth g ARC University of Aberystwyth www ATIBOOR ir 14 R Stelzer and K Jafarmadar 3 3 1 Austrian Society for Innovative Computer Sciences INNOC The team of INNOC focuses on control
113. cond half of November Once the trades are initiated they blow steadily from a northeast direction an optimal direction based on SOA s VPP data Increasing the survivability gale force winds are also rare in the winter months 1 www A TIBOOK ir 190 P Gibbons Neff and P Miller May 10 June 1 June 20 July 10 Aug 1 Aus 20 sept 10 Oct 1 Oct 20 Nov 10 Dec 1 Dec 20 110 100 90 80 70 60 J0 40 30 20 10 Number of Storms per 100 Years Hurricanes and Tropical Storms Hurricanes NOAA Fig 3 Atlantic Hurricane frequencies over 100 years NOAA showing the high proba bility of hurricanes in August October Fig 4 Hurricane paths from July September in the North Atlantic Ocean NOAA When leaving the Canary Islands the traditional objective is to leave the variable winds that surround the islands as quickly as possible A common recommendation is to travel on a southwest heading for about 1 000 nm This segment can possibly have northerly winds and will reach a turning point about 200nm northwest of the Cape Verde islands From there the boat will be just south of 20 N latitude and a course change to the west is taken This will ensure the boat reaches the northwest erly trades as soon as possible because they are rare to be found as far north as 25 N in the winter months 1 From December to March Force 6 winds are expected Since this is dangerous for the small SailBot the great circle route might be
114. ction augmented conceptual neighborhood this results in a changeover to relation pen After the turn it is a rea sonable strategy for A to move just about straight on which leads to relation 4X Going on like this brings us to relation 4x5 and ie subsequently At this point the rule could be considered to be successfully performed but to acknowledge that A should return to its original course we also add relation iG In summary we repre sent rule 12 a as the sequence denoted of formulae see also Fig 7 A 4x9 BAA 4Zi Ayina AB 4Zj Bwina with i 1 7 and je 9 15 2 A 4x B gt Agxt B A 4x B A 4x3 B A 4x4 B Finally based on such rule descriptions we define rule compliant behavior We assume rule R rj gt ra as given with x being the rule number Definition 1 Admissible configuration In context of a rule R r gt gt rn only configurations 7 included in the rule are admissible Definition 2 Agent behavior is rule compliant or admissible wrt R if 1 the vessels initial configuration is rp 2 during rule execution the vessels are only in admissible configurations 3 only changes from r to r occur and r _ does not occur after r 2 We are aware that this representation also includes situations where agents are far apart and no risk of collision is given For reasons of simplicity we do not to exclude these cases by a refined representation here www A TIBOOK
115. d behind are both conceptual neighbors of same In order to apply conceptual neighborhoods for reasoning about actions it is helpful to label neighborhood transitions with actions that initiate the respective transitions The resulting structure is called the action augmented conceptual neigh borhood 5 www ATIBOOR ir Rule Compliant Navigation with Qualitative Spatial Reasoning 145 behind Fig 3 Ordering relations arranged their conceptual neighborhood relationship 12 2 A 14 3 11 B 4 10 4 12 5 9 6 10 i 9 8 a A 443 B b A 423 B Fig 4 Two oriented points related at granularity m 4 3 1 A Qualitative Calculus of Relative Agent Position Navigation rules are often formulated in an egocentric frame of reference For ex ample the notion oncoming traffic refers to traffic traveling in the direction op posite to that of the observer In order to represent such knowledge we require a qualitative calculus about directional information We base our formalization on the OPR Am calculus which describes relations between object in the domain of oriented points in the plane i e 2D points equipped with a direction OPR Am re lations describe the position of an object B as seen from A and simultaneously the position of A as seen from B Relations are thus pairs of directions written A m lt B where i j denote the directions The calculus is defined for an arbitrary granularity m N controlling how many direc
116. d may only be required if the multi system powered test fails see below 4 Multi system powered test Additional loading to the power supply is pro vided by bringing up each subsystem that has passed the powered test one at a time and verifying that each maintains its performance and interoperability 5 In situ test The components are installed and again tested 6 Full system trials Here each capability of the system is tested as well as possible with variables removed Navigation tests are conducted on land mov ing the vessel on a cart This is followed by on water tests of basic GPS based navigation then maneuvering tacking etc and so forth building up sophistication www A TIBOOK ir 94 B E Bishop et al When students follow the design process they will inevitably run across a sub system that does not meet specs This is where the iterative nature of the design becomes apparent and where all of the preliminary work pays off Students have fall back options in place for any subsystem that fails and can refine decisions and metrics based on experience in the testing phase Without this careful and sys tematic approach students are often left accepting less than desirable performance because they simply cannot determine where things have gone awry and in most cases are not prepared to generate an alternative solution even if they did under stand the primary failure mechanism 1 3 Pedagogy of Robotic Sailing
117. d using hot wire cutting technique which is widely used in RC plane model construction The design core board extruded polystyrene enables fast and robust prototyping of any type of boats The polystyrene can be made more resistant by covering it with a fiberglass layer We decided not to actuate the sail to increase even more the robustness of the boat The sail is thus set at a fixed angle an empirical angle which works the best However even if we loose optimality the preliminary tests on the Ty Colo lake see Figure 8 showed that the boat can navigate upwind The navigation between waypoints as shown in section 3lhas also been tested using the prototype www A TIBOOK ir Sailing without Wind Sensor and Other Hardware and Software Innovations 37 f safety buoy l f i j gt M pees oo lead gt _ i ii So K electronics 5 Bl i keel aluminum board Fig 11 The core board of L improbable is an aluminum board which supports the mast the rudder and the hull 5 Conclusion In this paper we presented a possible solution to reduce electrical power consump tion by replacing electronic boat heading correction electronic autopilot by a me chanical correction using a wind vane self steering device We presented simulated results of heading stabilization and navigation maneuvers using the self steering de vice Simulations look promising The advantage of navigation using only
118. de data such as identification position and other critical information from nearby vessels within radio range The data is also forwarded via Iridium satellites to Data Assembly Centers where it is validated and safety relevant messages and warnings are distributed to the remote stations Since 2000 the International Maritime Organization s IMO International Con vention for the Safety of Life at Sea SOLAS requires that every registered vessel over 300 gross tones and all passenger ships regardless of size have to carry an AIS Therefore more than 40 000 ships are currently equipped with AIS 2 2 RoboCup Soccer Server Simulator The RoboCup offers various robot competitions e g RoboCup Soccer RoboCup Rescue and RoboCup Home Another contest is the RoboCup Soccer Simulation League where the focus is on the artificial intelligence and team strategy during virtual soccer matches without real robots For this competition a multi agent system was developed the RoboCup Soccer Server Simulator 7 www A TIBOOK ir 160 N Ammann et al This system enables two teams of eleven simulated autonomous robotic players and one coach per team to play soccer Simultaneously the server provides moni toring and logging of the matches as well as judging the rules For this reason the server was built as a client server system which is able to handle 24 clients Ev ery client is a single process or program and is connected over its own port to t
119. deo movies from the USV system the first frame can be seen in Figure lO Fig 10 Real time Video Algorithm Performance Real video camera on USV in different light and sea states and targets radius Selected movies can be found on line at 13 www A TIBOOK ir Automatic Obstacle Detection for USV s Navigation Using Vision Sensors 139 Target recognition and identification marked with a red rectangular and can be found at http sites google com site orenusv home research I usvvision The parameter values in our simulation are 10 the lower threshold value was set to 20 and the upper value was set to 100 the surrounding rectangles size were set to A 7 B 16 and fac 3 4 The tested scenes contained different sea states at day and night with different light conditions The target radius changes between 10 60 m The horizon line was examined with and without buildings in the background of the USV The run time for average on frame was measured as 0 005 millisecond with the minimum value 0 003 millisecond and the maximum value 0 014 millisecond Based on these results the algorithm can be used for real time application for au tonomous movement of USV identifying targets around the vehicle 5 Discussion and Conclusion This paper presents a basic and very efficient algorithm for marine environments that was tested on different scenes Yet there are some limitations and untested cases that should be treated in fu
120. devices A source arrow indicates a device that actually sends data to that multicast group A sink arrow in dicates a device that uses IGMP to subscribe to that multicast group Data streams that have no sink are probably used to communicate with extensions we did not in stall Streams where the same device class is source as well as sink are presumably for internal coordination between multiple devices of the same class e g multiple display units From the five multicast streams that connects the radar and the display unit in one or the other direction we identified two operation critical ones and tried to name them The Image Data Stream and the Control Register Access 4 Wireshark iS a network protocol analyser Project website http www wireshark org gt As it turned out any IGMP unaware switch will also work These switches will handle multicast traffic like broadcasts and therefore retransmit it on all active ports 6 The bridge module http sourceforge net projects bridge is avail able in the official linux kernel distribution www ATIBOOR ir A Digital Interface for a Lowrance Broadband Radar 173 236 6 7 4 6768 224 2 1 2 2048 236 6 7 18 D 236 6 7 8 6673 y F Radar Image Data a i i 236 6 7 9 6679 Do 236 6 7 10 6680 gt Register Control i 236 6 7 5 6878 a a Bi Ta ki r wu on gl Display Unit Ny 236 6 7 19 6689 D C 236 6 7 20 se00 F A sends UDP Datagrams w
121. dspeed Sail Winch ASS040 RC receiver PWM Failsafe modul Fig 2 System architecture of the control system Sensors left actuators right bottom Bluetooth and USB communication and power supply 2 PWM 6 PWM Controller j J Controller LM358962 CAN bus LM3S2110 UART2 I12C1 SPI1 UARTO UART1 I2C0 SPIO i2c Compass HMC6343 Fig 3 System architecture with CAN bus compass separated from the main controller Table 2 shows the main characteristics of the controllers and the number of the dif ferent interfaces The controller needs a supply voltage of 3 3V All peripheral pins are 5 V tolerant that means we dont need to change the logic levels of 5 V peripherals For the boat we use an evaluation kit EK LM3S6965 from TI which provides the following additional components In circuit debugging interface USB interface www ATIBOOR ir Using ARM7 and uC OS II to Control an Autonomous Sailboat 105 Table 2 Main characteristics of the Stellaris Cortex M3 controllers and the available inter faces Controller LM3S6965 without CAN LM3S8962 LM3S2110 CAN Main clock 50 MHz 50 MHz 25 MHz Flash RAM 256 KB 64 KB Timers 4x 32 bit or 8x 16 bit SysTick Watchdog PC 2 171 SSI SPI 1 1 1 UART 3 2 1 PWM 2x3 2x3 2 Ethernet 1 l CAN 1 1 communication over UARTO programming of the device debugging SD card slot buttons status led and an 128x64 OLED display Figure 4 shows the contro
122. e Rising edge Save the time since the last rising edge W_ Dir Period OxFFFF TimerValueGet TIMERO BASE TIMER_A Reload the timer TimerLoadset TIMER BASE TIMER A 0xFFFF Fig 9 Interrupt handler for port D Detect changes on BIT4 PWM signal of the AS5040 the heading directly over the I2C bus with a default rate of 5 Hz So we can get new values every 200 ms Figure shows the code for a task reading the data of the HMC6343 once a second OSTimeDly ensures that the task sleeps for 1s see line 13 The tasks may be prioritized depending on the requirements The www A TIBOOK ir Using ARM7 and uC OS IH to Control an Autonomous Sailboat 111 0l starcie voici Apoo TaskIr2 voice 0 arg 02s 4 03 CPU_INTO8U err OA CPU_INT16U heading 05E 06 while DEF_TRUE 07 get the heading of the boat and display it 08 heading BSP_I2C_Read_HMC6343 09E OSSemPend LCDSem 0 amp err MOE sprintf App_LCDLine9 I2C d heading Jede OSSemPost LCDSem Ze 13 OSTimeDly 0OS_TICKS_PER_SEC al L5 3 Fig 10 Task reading the compass data over the I2C bus once per second combination OSSemPend OSSemPost locks the display for this task to ensure a cor rect viewing of the value line 9 11 The exclusive access to a shared variable is possible in the same way Shared components in this system are the display and the Bluetooth UART They are locked with semaphores 4 3 Control Software
123. e ostasis within a sailing robot would enable it to modify its behaviour in accordance to varying conditions For example taking the analogy of the hormone insulin which controls the uptake of glucose from the bloodstream An artificial insulin could be used to activate power consuming behaviours such as switching on oceanographic sensors using communications equipment or making more frequent or larger course adjustments when battery levels are plentiful When battery levels are low these be haviours can be suppressed to reduce power consumption The robot s activities could also be promoted and suppressed in response to diurnal or seasonal solar cy cles or tides so that power consuming activities are coordinated with the presence of favourable conditions Getting BeagleB to achieve the long missions required for this research is a non trivial task Due to the financial cost of the robot and less than perfect reliability record of the onboard electronics and software we are currently unwilling to allow it to operate for prolonged periods without supervision from a chase boat or to op erate it in heavy seas Achieving this time commitment especially when combined with the overheads of launching and recovering BeagleB has been difficult and has severely limited the amount of time for on the water experiments Simulations offer the ability to run many iterations of these experiments without the need for chase boats and chase boat crews They also
124. e alf is made of fibreglass and carbon and there fore is relatively lightweight Boat has a length of 2 4 m and a height of 3 m It features 1 5 m of sail area in a combination of two sails main sail and jib both mounted on a balanced rig Used sensors are an electronic compass a wind sensor speed and direction and a GPS receiver The sensors are connected to a microcon troller via CAN bus 14 Boat competed in Microtransat 2006 and 2007 as well as in WRSC 2009 3 3 4 Queen s University Mostly Autonomous Sailboat Team MAST was founded in 2004 at Queen s uni versity Their first vessel entering SailBot 2007 was Black Adder Figure Tall a 2 m long carbon fibre hull with traditional sails using a PBasic Stamp for the con trol system Since 2007 the team of undergraduate students made significant mod ifications to their first design and participated in the Microtransat Challenge 2007 WRSC 2008 and 2010 as well as SailBot 2008 2009 and 2010 3 3 5 Swiss Federal Institute of Technology Zurich ETH Avalon Figure Sal was developed by a team of students from the Federal Insti tute of Technology Zurich for the Microtransat challenge and participated in WRSC 2009 Avalon features amonohull design with a length of 3 95 m a balanced rig and a twin rudder system The power supply is realized with four solar panels of 90 W four lithium manganese batteries of 600 Wh each and a direct methanol fuel cell for back up power The con
125. e a far superior life in comparison to current engines However considering a voyage across the Atlantic for about five months the rudders being controlled only a tenth time to save energy we can expect that we will use the motors for a period between 300 and 400 hours e Waterproofness must be taken into account on a sailboat because it would be faced with a marine environment during the Atlantic crossing The lack of wa terproofness may force us to realize a complex system to prevent water from entering the motor e To calculate the torque we need to determine the movement of the cables which control the rudder during the rotation for an angle of 35 degree and the force on the rudders The calculated nominal torque taken into account for the rudder is then 127 8 N cm i e 12 78 Kg cm e The DC motors having no common current limiter can withstand peaks equal to 5 times their rated torque However the peaks create an overheating of the mo tor Brushless motors with a current regulating canSt withstand overload greater than 50 of their rated torque According to calculations the maximum torque supported by the motor is 504 9 N cm id 50 49 Kg cm e the Input voltage gives us information on the consumption of the engine The Table 2 clearly shows that despite their durability the BLS151 motors can t bear the torque generated by the rudders The RX 28 has the disadvantage of not be ing designed for marine environments and therefore requires
126. e crossing from St John s to Fenit yielded a predicted fastest time of 17 4 days a slowest time of 30 3 days and a most likely time of 19 4 days 4 1 2 Southern Route The second possible route is a southern route taking advantage of the trade winds The pilot charts indicate variable winds from northern Europe that can cause signif icant problems While departure points in Ireland UK and Portugal and Spain were evaluated the only route that compared favorably to the Northern Route started in the Canary Islands the traditional final departure point from Europe Numerous www A TIBOOK ir Route Planning for Micro transat Voyage 189 ee F Maximum Ice Limit iu i AT Fig 2 Pilot Chart detail for July for the area around Newfoundland NOAA routes with similar transit times exist but to minimize ship traffic the destination was selected as Antigua This is one of the oldest routes as Christopher Columbus sailed this route during his four voyages to the Caribbean His fastest time was 21 days 1 The route begins from the island of Tenerife With its larger size than the three islands to the west it is still close to the open ocean but has a greater logistical support Departure timing is as tight as the Northern Route The optimal time to leave the Canaries is the last two weeks of November Referencing Figure 3 by this time a late hurricane will be a rare occurrence Additionally the winter trades are seldom before the se
127. e significant if solar panels are the primary means of generating power Another approach used to evaluate the route selection was the Navy s Green Amber Red GAR approach This mission risk analysis tool qualitatively evalu ates each risk based on the operator s experience The number of Red or high risk activities drives the go no go decision Figure 6 shows the model applied to the two route choices In this case the Southern Route appears less risky 5 Short Term Planning While the Pilot Charts offer a good understanding of the expected average condi tions most people realize that average weather rarely seems to occur El Nino and www A TIBOOK ir 192 P Gibbons Neff and P Miller a Corner Broo NEWFOUNDLAND a TERRE NEUVI Port Aus Basques Fig 5 Traffic density around Newfoundland indicating a northeasterly course out of St John s will avoid the most traffic Fisheries and Oceans Canada Fig 6 Qualitative GAR analysis of the two route options La Nina conditions may dramatically impact average climate conditions in any given year Short term events such as gales may dramatically differ from the long term av erages and a successful autonomous voyage will require carefully understanding the short term forecast Using the same approach described for the Long Range rout ing commercial software will calculate optimized courses using Gridded Binary www A TIBOOK ir Route Planning for Micro transat
128. e similar to that used by the actual boat in order to allow for development of a common code base He presents results showing a close match between sailing the same course in the real world and in simulation Mathematical models of sailing boat movement 8 7 have also been developed although some effort will be required to bring these to the point where they can form part of a real time simulation environment 3 Methods This section discusses the BeagleB sailing robot the Tracksail simulator and an HIL simulator which combines Tracksail with a copy of BeagleB s electronics www A TIBOOK ir Simulating Sailing Robots 115 3 1 BeagleB Sailing Robot 3 1 1 Control System Overview BeagleB is a 3 65 m long Mini sailing dinghy its full specifications are shown in Table 1 Autonomous control is provided by a Gumstix single board computer run ning a Linux based operating system The Gumstix processes data from the wind sensor GPS and compass and determines appropriate actuator positions to sail its target course Low level control of actuator positions is performed by a PIC18F4550 microcontroller which receives target positions from the Gumstix A separate sys tem of 4 LEM CAS 6 NP hall effect current transducers are used to monitor the power consumption of the entire system One transducer is attached to each actu ator another to the output of the solar panels and a fourth to the battery output These produce a voltage proportional t
129. e test system for the control components right American Model Yachting Association www A TIBOOK ir Using ARM7 and uC OS II to Control an Autonomous Sailboat 103 Table 1 Basic parameters of the boats Boat FHsailbot Saudade Length overall 152 cm 112 cm Length waterline 148 cm 103 cm Beam 33 cm 26 cm Draft 81cm 26 cm Displacement 15kg 9kg Sail area 0 65 m 0 52 m 3 Control Hardware At the beginning of the project we have a long discussion about the controller needed to control the vessel Beside microcontrollers we use microprocessor for mobile devices like notebooks in different multimedia projects e g Pentium M In tel USA and similar processors and for small PC like systems with 386 486 cores On the other hand we use microcontrollers from Texas Instruments TI USA like the MSP430 for many small systems As a compromise between performance and power needs for this project we decided to use an ARM7 microcontroller Stellaris Cortex M3 from TI as the main controller Only its use is discussed in this paper see details later in this chapter The controller gets data from different sensors GPS 3 axis compass wind direction generates PWM signals for sail winch and rudder servo and communicates over Bluetooth An USB interface is used for pro gramming and debugging Status information may be displayed on a small OLED display Beside the internal FLASH memory additional data may be stored on a SD card Figure 2
130. e under the side constraints of rule consistency Our approach can be regarded as a hybrid navigation system involving mathematical models in the terminology of Statheros et al 15 which argue for the use of hybrid models in ship collision avoidance 3 Qualitative Spatial Knowledge Representation Qualitative Spatial Reasoning QSRJ is the subfield of knowledge representation involved with spatial representations that abstract from the details of the physical world Its reasoning techniques allow predictions about spatial relations even when precise quantitative information is not available 4 Based on qualitative represen tation and corresponding reasoning methods computers can be enabled to monitor diagnose predict plan or explain the behavior of physical systems 9 In gen eral two categories of reasoning based on qualitative spatial representations can be distinguished constraint based reasoning to reason about static configurations and neighborhood based reasoning to reason about how qualitative representations can l As reasoning is not possible without representation we will not distinguish between them generally in the remainder of this paper That is we shall refer to qualitative spatial repre sentation or qualitative spatial reasoning only www ATIBOOR ir 144 D Wolter F Dylla and A Kreutzmann change over time In the following we give an intuitive introduction to the basic concepts of QSR the interested reader
131. e will present in this paper is to improve the reliability of Breizh Spirit for the Microtransat challenge The design of the last two boats is the result of several studies based on the experiences of Breizh Spirit 1 and other boats In the first part we develop some points which were particularly studied Moreover electronics were also completely rebuilt to improve reliability Nevertheless Breizh Spirit 2 beyond the crossing of the Atlantic is also used for research programs to study the behaviour of the boat at sea That is what we will develop in the third part Finally we will present the results of our work at different validation tests www A TIBOOK ir Breizh Spirit a Reliable Boat for Crossing the Atlantic Ocean 57 Table 1 Boats Characteristics unit Breizh Spirit 1 Breizh Spirit2 Breizh Spirit 3 LOA m 1 5 2 3 1 7 LWL m 1 3 2 1 4 BWL m 0 35 0 8 0 45 T m 0 8 0 8 0 8 SA m 0 8575 2 0 75 Disp kg 13 55 13 Cb i 0 6 Fig 2 servo motors protec tion system In the Table 1 we characterize our boats by Length Overall LOA Waterline Length LWL Displacement Disp Waterline Beam BWL Draft T Sail Area SA and Block coefficient Cb 2 Development of a Platform Adapted for Crossing the Atlantic Ocean 2 1 Design of a Transatlantic Boat After the damages caused during the endurance test between Brest and Morgat we chose to draw conclusions from Breizh Spirit 1 to finalize the construction of Breizh Sp
132. eed It is desirable to have actuators that have no required holding torque using a mechanical brake or a non backdrivable system such as a worm gear Communication The system must communicate its state to a base station for monitoring and data collection and must allow remote reconfiguration Control The system must be able to take outputs from the processor and use them to manage the actuation Further there must be a remote control capability that is the system default that is when the processor fails the system automati cally defaults to remote control Sensing GPS wind data obstacle locations etc must be provided to the proces sor as needed for the appropriate algorithms Power An integrated power system must be generated that will provide appropri ate power to all components regardless of load This often includes multiple supplies as well as regulation to achieve optimal balance of weight volume and capacity Containment Mounting All system components must be securely mounted and protected from water incursions which are common in small amounts The com ponents must be easily accessible and must fit through the access panels on the vessel Most of the components used in USNA SailBot are commercial off the shelf COTS products from the R C sail winch to the AirMar weather station used for wind measurement and GPS data The full list of components for SailBot changes from year to year and is beyond the scope of th
133. ems Acknowledgements The author would like to thank Yaron Vazana for his help during this research and to Dr Blaurock and Dr Schlaefer for their helpful comments www A TIBOOK ir 140 O Gal References 1 Bhanu C Holben R D Model based segmentation of FLIR images IEEE Trans Aerosp Electron Syst 26 2 10 1990 2 Ben Yosef N Bahat B FeiginBhanu G Holben C Simulation of IR images of nat ural backgrounds Appl Opt 22 190 193 1983 3 Ratches J A Walters C P Buser R G Guenther B D Aided and automatic target recognition based upon sensor inputs from image forming systems IEEE Trans Pattern Anal Mach Intell 19 1004 1019 1997 4 Casasent D P Neiberg L M Classifier and shift invariant automatic target recognition neural networks Neural Networks 8 1117 1129 1995 5 Schachter B J Lev A Zucker S W Rosenfeld A An application of relaxation meth ods to edge reinforcement IEEE Trans Syst Man Cybern 7 813 816 1997 6 Walters D K W Computer vision model based on psychophysical experiments Pattern Recognition by Humans and Machines 2 1986 7 Broy M Early vision In Rosenfeld A ed Perspectives in Computing pp 190 206 Academic Press New York 1986 8 Marham K C Comparison of segmentation processes for object acquisition in infrared scenes IEEE Radar Signal Process 136 13 21 1989 9 Davies E R Machine Vision Theory Algorithms Practicalities
134. esign with a relatively inexpensive kit readily available Based on the kit and the description of our prototype we expect that it would be relatively easy to build an rrMM class boat Moreover despite its size the boat can be sailed in a fairly wide range of conditions of up to 5 Bft allowing to test algorithms on virtually every lake or in otherwise sheltered waters Our results regarding the sailing performance underline that the rrMM class presents a reasonable platform for development of robotic sailing methods The boat speed of upto 2kn allows to quickly cover a race course or to study some planning scenario particularly as the boat can sail relatively close to the wind Essentially if algorithms work with an agile and dynamic boat like the rrMM it can be expected that they could be ported to larger boats Moreover programming and testing is simplified by the option to use onshore computers for boat control Although this implies the boats are no longer strictly autarkic they are still autonomous in the sense that they are fully computer con trolled Hence the new class presents a viable alternative for groups interested in testing new approaches particularly for multi boat scenarios However we do not see the rrMM class as a replacement for the SailBot and Microtransat classes Instead it complements these classes both in presenting an affordable and small entry level platform and by focusing on algorithmic aspects The latter also
135. etector algorithm computing the gradient intensity at each point x y in the frame I x y V Vx Vy 6 __Vy J 7 0 tan Vx 7 www A TIBOOK ir Automatic Obstacle Detection for USV s Navigation Using Vision Sensors 131 Fig 2 Initial Cleaning of the Image The left side shows the original input image after the transformation into a gray scale format The right side shows the output image of the Canny Edge Detector Figure 2 shows the original input and the output image of the Canny Edge De tector The next step is to reduce the search space using a simple Hough Line Trans formation to get the most intense line in the image 3 3 Learning Sea Pattern This part is the actual recognition part of the algorithm We assume that there are no targets within 10 pixels from each edge of the frame This assumption is legitimate based on the fact that the input device is usually a video device which can be easily adjusted and focused on the center of the frame Learning the sea pattern can be done by an occurrence matrix The occurrence matrix is N x N where N is the possible number of gray levels inside the frame 9 We define a length function f from current pixel x y f x x ly 8 where lg is the distance to the proportional pixel Eq 8 is used to determine the occurrence matrix values depending only on x values based on the sea texture as suming that the sea texture is horizontal In the next stage we scan the
136. expanded After a series of node expansions the set of active nodes is sampled to cut it down to its initial size Doing so the memory requirement of the planner is kept constant We perform the sampling simply by first selecting all k nodes to ex pand as explained above and then performing node expansion Nodes that have not been selected are discarded immediately Although limiting the set of active nodes may discard states that lead to the goal the step is necessary to obtain a method that can generate a plan using limited computational resources In our evaluation we analyze different choices for the size of the set of active nodes in order to identify a good balance between the ability of the planner to determine a path and memory requirements www ATIBOOR ir 150 D Wolter F Dylla and A Kreutzmann symbol value description Exil 50 size of sail in m Cm 10000 mass in kg Chu 0 02 coefficient longitudinal vs lateral friction of hull Cup 0 5 buoyant force Carift 0 001 wind drift Cfr 0 005 friction Crot 100 rotational inertia Cr 5 0 rudder force coefficient Fig 8 Legend of variables and parameters used in the physical simulation 5 2 Physical Simulation Designing our simulation we aimed to create a mock up of the sailing experience with cruising yachts approximately 10m of length deep single fin keel and a sin gle mainsail As control commands we only consider the position of the rudder and the length of the s
137. f a reliable power monitoring system and to compare the results of power consumption estimates from Tracksail and actual recorded data from the HIL simulator and Tracksail The level of noise in the HIL system also needs to be varied to see if it is possible to recreate results seen on BeagleB e g the noise which caused multiple upwind tacks to be required in figure 4 Noise could also be applied to the simulated sensors in Track sail and this might give an increased level of realism to experiments run within Tracksail Acknowledgements This research was funded by EADS Foundation Wales as part of the grant Tethys References Sail simulator http www sailsimulator com accessed June 10 2011 Virtual sailor http www hangsim com vs accessed June 10 2011 Tracksail ai 2009 http sourceforge net projects microtransat files tracksail AI tar gz download accessed June 10 2011 4 Briere Y Iboat An autonomous robot for long term offshore operation In Proceedings of MELECON 2008 pp 323 329 2008 5 Crapo L Hormones The Messengers of Life W H Freemany and Company New York 1985 ISBN 0 7167 1753 0 oat hae oe www A TIBOOK ir 124 C Sauz and M Neal 6 Kuusela T Brockmann S Johnson R Tracksail 2004 http tracksail Ta 10 11 sourceforge net accessed June 10 2011 Ridder E J Vermeulen K J Keuning J A A mathematical model for the tacking ma neuver of a s
138. f action when we see a situation that the mariner may not have seen and e DIRECTING the outcome of situations when necessary to prevent disasters In order to achieve the goals mentioned above AIS provides these services with information such as unique identification position course and speed of observed vessels AIS was originally developed as a ship to ship navigational aid but evolved into a global standard Figure 1 shows an overview of the system which works as follows 6 The participating vessels automatically broadcast their information at regular intervals via an AIS Very High Frequency VHF Data Link transmitter This information can be received by other vessels or stations For example a fleet of buoys or stations with www A TIBOOK ir Collision Avoidance Using a World Server 159 a z 32 NIS AIS BUOY AIS BUOY u VESSEL DATAASSEMBLY ENTER ww sse DATA MONITORING AND VALIDATION Fig 1 This figure shows the components of the AIS the participating vessels broadcast their information via VHF so other vessels and stations can receive these data The stations AIS buoys forward the data via Iridium satellites to the Data Assembly Centers where the data in validated and safety relevant messages are sent back to be broadcasted Furthermore the data is made available on the Internet low power AIS receivers can be located offshore or along the coast These stations provi
139. f low sunlight For the proposed crossing of the Atlantic tropical road such days should be rare However in the case of a loss of battery power a charge controller can turn off the electronics and actuators power A memory system allows the boat to restart when the battery voltage reachs sufficient capacity to ensure proper operation During these moments without power the boat will behave like a drifting raft GPS position sent by the SPOT system has for security reasons a separate sup ply by a lithium battery to ensure the proper operation of sending the position three times per day for 6 months For this the SPOT casing has been modified to a mi croprocessor to wake him up only when it must send messages The entire equipment is housed in a waterproof case Links to other elements solar panels sensors microprocessor and memory programming way points pass www A TIBOOK ir 64 R Leloup et al Initialisation Reading the compass value Reading the GPS value Reading the buoy position Is the boat near the buoy yes no Reading wind 4 direction Calculation of the direction of the boat Adjust the rudder and the sail Fig 7 Navigation algorithm Validation Validation countert counter lt 4 Validation counter Next buoy through a series of waterproof connectors To ensure a complete waterproofness a magnetic control system replaces a possible waterp
140. f uC OS I period of the PWM signal W_Dir_Period line 21 and the time the signal is 1 W_Dir_Value line 15 The wind direction variable is updated every 1025 us Fig 8 right This variable may be read by different tasks which need the wind direction for internal calculations Other tasks read sensor values over different interfaces when needed Fig 8 left Corresponding to we can calculate the error corrected value of the wind direction in degree On the other hand the compass HMC6343 is able to deliver www A TIBOOK ir 110 M Koch and W Petersen Task1 Main control k Winddirection Task2 GPS AS5040 Task3 12C compass Interrupt Task4 Communication oa 1 ms Interrupt Service Routine Shared Variable Winddirection Fig 8 Tasks reading sensor values left vs interrupt driven calculations right 01 02 s OS 04 05 06 Om 08 09 IOE je IPE L3 s rar IESE Ge L72 L8 e 19 2 POE Ole 22 static void BSP _GPIOD IntHandler void CPU_INT64U intstat Get GPIO interrupt source intstat GPIOPinIntStatus GPIO_PORTD BASE true Clear the GPIO interrupt GPIOPinIntClear GPIO_PORTD BASE GPIO_ PIN 4 Interrupt on Pan 4 Wind direction if intstat amp GPLO PIN 4 if GPIOPinRead GPIO PORTD BASE GPIO_ PIN 4 0 Falling edge Save the time since the rising edge W Dir Value OxFFFF TimerValueGet TIMERO BASE TIMER_A els
141. follows from the class rules which limit the differences between boats Yet the proposed rules allow for some flexibility e g in principle permitting swing or wing rigs While we expect the rules and the design to evolve over time we hope that other groups will adopt the rrMM for their projects References 1 Alves J C Ramos T M Cruz N A A reconfigurable computing system for an au tonomous sailboat Journal of the Osterreichische Gesellshaft fiir Artificial Intelligence Austrian Society for Artificial Inteligence 27 2 18 24 2008 2 Ammann N Biemann R Hartmann F Hauft C Heinecke I Jauer P Kr ger J Meyer T Bruder R Schlaefer A Towards autonomous one design sailboat racing navigation communication and collision avoidance In 3rd International Robotic Sailing Conference Kingston Ontario Canada Ammann N Biemann R Hartmann F Hauft C Heinecke I Jauer P Kr ger J Meyer T Bruder R Schlaefer A pp 44 48 2010 www ATIBOOR ir 84 10 11 12 13 A Schlaefer et al Briere Y Bastianelli F Gagneul M Challenge microtransat In CETSIS Nancy France 2005 Bruder R Stender B Schlaefer A Model sailboats as a testbed for artificial intelli gence methods In 2nd International Robotic Sailing Conference Matosinhos Portugal pp 37 42 2009 Giger L Wismer S Boehl S B sser G A Erckens H Weber J Moser P Schwizer P
142. from which the currently best node is chosen for expansion In contrary to Dijkstra s algor thm the best node is not the node having the lowest cost but the node having the lowest value c x h x where h x is the value of the heuristics for the node x This heuristics gives an estimate for the cost to reach the node Xgest from x We use the distance from X to Xgesr divided by the boats hull speed as heuristic function h x It is consistent as it satisfies the triangle inequality h x lt ti h x for two adjacent nodes x and x and also the condition h Xgest 0 We choose this heuristics as it is cheap to compute and its consistency guarantees the optimality of the A algorithm 5 To find an optimal route the A algorithm is run on the routing graph described in Section 2 1 3 As the number of nodes and edges in the routing graph can get very high we tried to optimize the graph construction to save memory and computation time in the following way Some nodes in the open list do not need to be expanded if the destination node is reached before they are considered Also the fact that our heuristics is consis tent guarantees that a node will never be expanded more than once by the A algorithm 5 This allows to dynamically construct and deconstruct the routing graph during the execution of the algorithm e A node x is not created until one of the eight direct neighbors x in the grid is chosen for expansion e Edges are
143. g custom built hull and has a total sail area of 3 7 m The design was inspired by modern racing ocean yachts The control system is implemented on a small FPGA based single board computer in cluding a 32 bit RISC microprocessor running at 50 MHz Communication with the boat is possible using Wi Fi GSM Iridium SBD and a conventional RC receiver used in radio controlled models FASt entered in WRSC 2008 and 2009 4 Potential Applications Recent events like the devastating tsunami in Asia in 2004 the Deepwater Hori zon oil spill in Gulf of Mexico in 2010 accidents with refugee boats off the coast of Lampedusa Italy and pirate activities in the Gulf of Aden have emphasized 28 Colin Sauz 22 Colin Sauz 30 Alexander Schlaefer 31 Alexander Schlaefer 32 Jos Carlos Alves www A TIBOOK ir History and Recent Developments in Robotic Sailing 19 impressively the importance of a fully integrated ocean observation system 39 AUVs and motorised ASVs are widely used for ocean observations since many years 10 39 According to Cruz and Alves the main strengths for unmanned autonomous sailing boat for this task are Long mission ranges Negligible operational costs Potential for towing sensors Real time data transmission Real time localisation Very low noise generation Not all of the possible applications mentioned in the list below are likely to be realised in the next few years The current focus is clearly set
144. gn class we can report that handling the boat is indeed very simple We fre quently carry up to four fully rigged boats in one car and the boats are small and lightweight enough to fit into standard luggage The boats are also very robust on shore and while sailing We have sailed our boats in winds from 2kn to 20kn Overall the electronics are also quite stable with GPS and bluetooth active even when the boat is heeling The biggest issue is the compass sensor which even after a calibration as recommended by the manufacturer shows systematic errors More importantly the sensor seems to be sensitive to fast rotations e g when the boat is heeling or sailing in short waves Figure 7 summarizes compass versus GPS plots for four different boats yellow green red and blue Clearly surface currents and wind do have an impact as does the delay in the GPS heading but the plots illustrate Yellow oO 19 c9 Q Q _ 8 _ D D amp Sg D D Eo So TO TO oN oN oO oO c o 2 a g g E E O O 2 3 oO o O 50 100 150 200 250 300 350 O 50 100 150 200 250 300 350 GPS heading deg GPS heading deg Red Blue 2 2 Q Q n n BD BD oO B B D D So So TO TO oN ON o o N w N N i 5 Q Q 52 52 oO oO Q Q LO Ww oO oO 0 50 100 150 200 250 300 350 0 50 100 150 200 250 300 350 GPS heading deg GPS heading deg Fig 7 A comparison of the compass heading with the GPS heading for all four boats in s
145. greatly simplifies its design A mechanical design proposition will be provided for this device Our first aim is to validate trajectory stabilization by the device The final purpose is to sail using only the self steering device without any rudder on the stern Sailing using the self steering device enables navigation relatively to the wind As a consequence there is no need for the wind sensor Simulations are provided for navigation maneuvers such as tacking and jibbing for both cases when the regulator is on the bow and the classical case when it is on the stern In this paper we also present algorithms for navigation between waypoints There are many different methods which can be used for autonomous navigation 11 6 Our algorithm allows the boat to navigate on straight lines between waypoints taking the drift of the boat into consideration Simulations results are provided to prove the usefulness of the approach Finally we will present our test prototype which is used for testing the concepts above The design has been driven by the simplicity in construction 2 Sailing without Wind Sensor 2 1 Wind Rudder Regulator According to 4 the idea that a sailing boat might be able to steer itself using wind did not appear until the twentieth century It might be basically because the crew was cheap and was doing the job perfectly The first model of self steering wind vane was installed on a motorboat in 1936 It was connected to the rudder
146. gth 1 byte Setting 1 increases the scanning speed while O resets to normal speed 6 2 8 Register AO Keep Alive Payload length 1 byte The radar turns off automatically after 20 60 seconds when it loses contact to the display unit To prevent this a keep alive timer has to be reset regularly Register OxA0 has to be set to value of 2 every approx 10 seconds 7 Sample Data Acquisition Setup By May 2011 the mounting brackets for the ASV Roboat are still in construction To acquire real world sample data in a dry test we chose to mount a test setup of the system on a bike trailer The setup see figure 5 consists of radar antenna radar interface box a 12 V lead battery radar controller unit and a laptop For ease of use the laptop has been mounted on the front carrier of the bike and connected to the other components on the trailer On the laptop is run 1 the authors JAVA imple mentation to control the radar and check the output and 2 Wireshark to capture network traffic for later analysis To acquire positions a smart phone running a GPS tracking software has been mounted alongside the radar antenna www A TIBOOK ir 178 A Dabrowski S Busch and R Stelzer Fig 5 The sample data acquisition setup This setup allowed the authors to record sample data for dynamic values of speed and direction For further analysis radar data will be mapped to the recorded GPS data and fed into the mapping module of the Roboat software
147. h between large and small movements Further data is required before we can conclusively demonstrate that either Tracksail or the HIL system are reasonable approximations for the power consumption of BeagleB 5 Conclusions Through this work we have shown that simulators can be useful for testing and development of sailing robot control algorithms Even simulators with limited real ism and simplistic physics models such as Tracksail can provide a valuable aid to the process of control system development and debugging and to act as a parame ter verification method for scientific experiments investigating power management However even with an improved physics model no simulator is likely to be a full substitute for testing in a real environment Capturing the required level of detail to represent the sheer complexity and dynamic nature of the world encountered by a sailing robot would be a monumental task Given this simulation cannot provide an alternative to deploying a sailing robot in the real world but it maybe useful to the development process The use of Hardware in the Loop simulation provides a further aid to software and electronics development by offering a identical or almost www A TIBOOK ir Simulating Sailing Robots 123 identical platform to that which will be deployed at sea It can create an environment in which it is easy to instrument the system to observe electrical characteristics or software performance 5 1 Future
148. hanged e g to add openings 3 The keel fin and rudder must be from the Graupner kits either MicroMagic or racing MicroMagic 4 The material shape and weight of the keel bulb may be altered as long as it is not harmful to the environment 5 The jib and main sail may not exceed the area of the Graupner kit sails 6 The type of rig is open as long as the limitation regarding the sails is fulfilled and the boat fits into a sphere of 1 m diameter excluding sensors antennas mounted on the mast top 7 The boat must be human controllable throughout the races 8 The boat must communicate its true GPS position to a central server at a fre quency of at least 1 Hz 9 Boat control shall be completely automatic without human intervention These rules try to trade off the objectives of having directly comparable boats and allowing some room for improvements To achieve the former hull keel and rudder are rather tightly restricted and the sixth rule intends preventing too extreme rig designs However controlling the sails is a key problem in robotic sailing and under www A TIBOOK ir 74 A Schlaefer et al Table 1 Dimensions of the Graupner Micro Magic Note that the weight depends on the edi tion standard vs carbon and on the additional components to control the boat Our standard rrMM boats weigh approximately 1030 g with the standard keel and including the battery Dimension Value Length 530mm Beam 180mm Height 980mm Weigh
149. he server via UDP IP sockets Depending on the current situation every client receives continuously its own match information i e the server provides sensor data for the 22 player clients Moreover the coach client receives the positions and move ments of all players to adapt the team strategy Every 100 ms the server receives the player client s commands and computes the new match state Furthermore the server provides the connection of several monitors to observe the match and to change the server or match properties The RoboCup Soccer Server Simulator is under constant development and has been in use since 1997 3 Methods In this section we present requirements and criteria leading to the design of a World Server for robotic sailing First we consider the topology of the communication nodes and second we compare different communication methods 3 1 Topology of the Communication Nodes Our boats do not have sufficient sensors to detect other boats and obstacles on their own and they are also limited in resources like computing power Therefore every boat and obstacle must communicate their existence and position to the other boats in order to be detected Based on this information the boats have to plan a collision free path There are basically two options to realize the communication broadcasting like implemented in AIS and a centralized approach as used for RoboCup These two approaches are discussed below 3 1 1 Broadcasting
150. he Fill Filter For each white pixel target pixel if all of its eight neighbors are black sea pixels we change this pixel value to be a sea pixel The Cleaning Filter Matrix Cy is 000 Cyli j 010 10 000 The third filter is called Connection Filter In some cases Fill filter and Clean ing filter can not recover damaged pixels if the problem occurres in more than one adjacent pixel Therefore the Connection filter detects two separate components of a target and connects them The Connection Filter Matrix Cng is 001 CGyl j 101 11 100 Fig 4 Morphologic Cleaning The left side shows the image after detecting the sea pattern and the right side shows the image after applying the morphologic methods Cleaning and Fill filter www A TIBOOK ir 134 O Gal The forth filter is based on the assumption that each pixel value is determined by its neighbor pixel values This filter is called Majority Filter For each pixel is calculates the eight neighbor values and changes the current pixel value according to the majority The left side of Figure 4 shows the image after detecting the sea pattern and before applying filters The right side of Figure 4 shows the image after applying the cleaning filters We can see that all the damaged pixels were removed and we managed to derive a clean target image 3 5 Marking Interesting Points The video device and the detection algorithm are not perfectly accurate and dam
151. he naviga tion problem in autonomous robotic sailing Formalizing spatial knowledge occur ring in sea navigation essentially involves representation of directional spatial infor mation i e to describe the positions as seen from specific points of view egocentric frame of reference Our approach employs relations from the qualitative constraint calculus OPRA according qualitative reasoning methods allow us to combine information from different frames of reference into a coherent whole Qualitative directional relations can capture static traffic regulations as imposed by buoys as well as it can capture spatio temporal movement patterns of for example official right of way regulations or strategic maneuvers While qualitative reasoning can be used to determine coarse qualitative actions that are admissible with respect to the navigation rules additional means are required to check whether such actions are possible for a specific agent in a specific physical context In particular the kinemat ics of sailing vessels largely depend on the current wind speed etc We use prob abilistic roadmap planners to determine applicability of actions The randomized approach to planning is particularly attractive for its ability to cover large search spaces Furthermore the approach can easily be integrated with a qualitative rule formalization In this paper we demonstrate how the integration can be achieved We also give first results of an integrated ra
152. heet rope that controls how far the boom is opened We employ a very efficient but idealized physical simulation to determine the effects of ac tions and development of the environment Most essentially we make the following idealization e No waves no fluid simulation e Simple wind model no turbulence no slip streams e Control actions are performed in a single simulation step Although true sailing sport draws its attraction from some of these facets we believe that they can be neglected in context of navigating in safe operation range Unfortunately any realistic simulation involves careful modeling of physical phe nomena beyond the scope of our work and it would require considerable compu tational resources Since roadmap planners make intense use of the simulation its efficient implementation is key We employ the variables shown in Fig 8 to describe the dynamic physical state of any vessel By v we refer to the left normal vector of a vector v and v ands for the lee site normal vector In order to model the water resistance we decompose friction into friction along its longitudinal and its lateral axis LL st d f h a f h h B h h 5 Based on the current local wind vector w obtained as the difference between global wind and current speed we determine the resulting acceleration which is then integrated in a constant time step simulation to update the dynamic vessel parameters www ATIBOOR ir Rule Co
153. hing be tween sea and possible targets 6 Skeletonize the recognized targets in order to characterize them with a sim ple structure 7 For each target Lower the characteristic structure into the interesting points according to the target skeleton structure that was found NO me Basic methods in image processing enable to indicate targets position at a spe cific frame with a small effort Naive algorithms report target s current parameters immediately Instead we analyze the 20 input frame at each time to find possible targets We count the input frames and treat only the current 20 input frame com paring and correlating the results with the last 20 input frame We decided to use the 20 input frame as optimal frame number based on the parameters of minimal www A TIBOOK ir Automatic Obstacle Detection for USV s Navigation Using Vision Sensors 129 Flow Chart REENE Learning the Sea Marking i Pattern Interesting Points Reducing Search Morphologic TE A S pace Cleaning Time Correlation Output Blurring Filters Thinning Targets Fig 1 Flow Chart of the Algorithm Describing algorithm s functions from input frame to algorithm output using major steps Reduce search space Learning sea pattern Morphologic cleaning and Skeletone algorithm target changing in marine environments and CPU time computation We analyzed these parameters on our real time video movies with several frame numbers and decided to
154. hitecture and physics electrical engi neering and power managment embedded systems computer science and systems engineering Establishing a conference held jointly with the World Robotic Sailing Champi onship WRSC has provided a platform for discussions among scientists from all fields involved in robotic sailing In fact we believe that the progress made in au tonomous sailing so far is to no small extent driven by this combination of compe tition and knowledge exchange The interdisciplinary nature of robotics and robotic sailing is reflected in the papers contributed to IRSC and the teams participating in WRSC Further promoting this multidisciplinary approach will be key to tackling the numerous challenges on the way to truely autonomous sailboats These proceedings summarize the state of the art in robotic sailing and the in troduction in Part I contains a review illustrating its history and recent advances Clearly having a robust and reliable boat is a key requirement which is also the fo cus of papers in Part II The proposed designs range from small one design boats for algorithm development to vessels built to cross the Atlantic Ocean Different aspects of the system design and validation are discussed in Part Ill The remaining papers focus on algorithmic matters Part IV presents approaches for collision avoidance while Part V addresses localization and route planning www A TIBOOK ir VI Preface Organizing IRSC 2011 wa
155. i nates relative to the vessels true heading and position That way the head up radar data is transformed north up As position and heading information is only available at much larger intervals than the radar scan lines position and heading will have to be extrapolated based on current course over ground and turn rate Instead of extracting and converting obstacles from polar radar to a Cartesian map that again produces polar data a hybrid algorithm may be used Map and radar may each produce a preferable directions polar diagram of their own and merge them in a final step This approach involves less complexity but does not allow to use dynamics of obstacles and therefore calculations like the time and distance of closest approach as needed for some COLREG decisions 3 section 4 2 11 Criticism and Conclusion The empirical found protocol described in this paper allows the direct digital ac cess to radar imagery on a low level from a Navico Broadband Radar via standard Ethernet and an UDP IP stack This is an easily available interface for mid range embedded systems This paper does not deal with any of the other issues involved in operating a radar and reading its imagery Especially it does not deal with auto matic adjusting e g gain preprocessing eg removing radar artefacts or analysing the data and extracting obstacles All topics we will have to face in the next months We have also not tested special cases like 6 where the boat
156. ic Sailing The Robotic Racing Micro Magic Alexander Schlaefer Daniel Beckmann Maximilian Heinig and Ralf Bruder Abstract A number of boat designs have been proposed for robotic sailing par ticularly inspired by competitions like the Microtransat challenge SailBot and the World Robotic Sailing Championship So far most of the boats are one offs often highlighting naval architecture aspects We propose a new one design class based on a readily available kit Small lightweight and with proven sailing performance the robotic racing Micro Magic presents a more standardized alternative particularly for algorithm development and multi boat scenarios Our intention is to introduce an evolving class and we propose a set of basic rules and describe the modified boat design electronics sensors and control approach for our prototype Moreover we have used four identical boats for the past year and we present results illustrating the good and comparable sailing performance indicating that the class is suitable to study robotic sailing methods 1 Introduction Competitions like the Microtransat challenge 3 SailBot or the World Robotic Sailing Championship provide environments for evaluating and comparing the over all performance of robotic sailboats As of now a number of small and medium sized boats have been built by different groups However with the exception of the mobile ocean observation platform MOOP designed by a team from Aber
157. ic solar panels and leaves a power budget which is only just sustainable for long term operation at moderate latitudes winter time operation at extreme lati tudes would still not be possible Given these limitations power must be managed efficiently To achieve this a biologically inspired mechanism based upon the mam malian endocrine system has been implemented 11 The endocrine system operates by secreting chemical messengers called hormones into the blood stream 5 These spread throughout the body in a broadcast fashion however they only effect certain target cells When they encounter a target cell the hormone will bind with receptors on the surface which are shaped only to fit that particular hormone After this binding occurs the hormone will trigger a behavioural change within the internal workings of the target cells Hormones are responsible for the regulation of a number of biological processes including sleeping patterns blood sugar lev els salt levels blood pressure coordination of actions between the neural immune Fig 1 A Photograph of BeagleB during the 2009 World Robotic Sailing Championships in Matosin hos Portugal www A TIBOOK ir Simulating Sailing Robots 117 and endocrine system and triggering the fight or flight response which occurs in dangerous situations These regulatory properties combine together to form a stable internal state within the body known as homeostasis Creating an artificial hom
158. ich is an inter section of two other lines 1 e considering thin target for each white pixel Skeleton pixel we perform the Convolution Operator C with the matrix ca a2 The matrix Co identifies most significant points within a two line intersection with a predefined threshold S in the Skeleton algorithm As mentioned before CPU time is a major issue in our algorithm implementation Tracking targets in the next frames is based on a simple structure of the target reducing CPU time Moreover it is much more accurate to characterize a target with a few points rather than using a large number of points in such a clutter and noisy environment Figure 6 shows the skeleton image after the thinning process the left side demon strate the target interesting points implementation Fig 6 Skeleton Algorithm with Interesting Points The left side shows the target image after Skeletone thinning process and the right side shows the Skeleton algorithm with Interesting Point www ATIBOOR ir 136 O Gal 3 6 Time Correlation This is the final step in the recognition algorithm part After we have achieved two sequential frames and identified the targets in each one of them we apply the time correlation step Our goal is to make the final decision about a possible target that was found in the first frame and to decide whether it is a real target or a false alarm Given a set of interesting points in the first frame N n n and an
159. icinity of the server As this is no limitation for smaller boats and our typical race setup we have opted for a centralized World Server 3 2 Communication Method Regardless of the approach boats have to communicate Essentially there are two widely used wireless communication technologies we considered First wireless lo cal area networks WLAN provide high data throughput in a limited area Second enhanced Bluetooth modules can cover a similar area with considerably lower band width compare Table l However one important aspect in the context of small boats is power consumption which is typically far lower for Bluetooth Either choice would be open for teams connecting to the World Server and both do not require special permissions for operation 4 Implementation Our implementation of a World Server system consists of three major parts the boats clients a communication gateway and the actual world server It is a further development of our system introduced in 4 4 1 World Server Figure 2 illustrates an example setup of these components Note that in the ex ample the client is split into a gateway routing the communication to the World Server and the actual clients written in Java or C Clearly this could also be imple mented in a single piece of software The World Server collects data from all clients www A TIBOOK ir 162 N Ammann et al lt TCP sockets Fig 2 World Server components overview
160. ilboats In FUZZ IEEE 2007 pp 1 6 2007 17 Westphal M Dornhege C W lfl S Gissler M Nebel B Guiding the generation of manipulation plans by qualitative spatial reasoning Spatial Cognition and Computa tion 11 1 75 102 2011 www ATIBOOR ir www ATIBOOR ir Global Data Storage for Collision Avoidance in Robotic Sailboat Racing the World Server Approach Nikolaus Ammann Florian Hartmann Philipp Jauer Julia Kr ger Tobias Meyer Ralf Bruder and Alexander Schlaefer Abstract Collision avoidance is very important for autonomous sailing with many boats e g during races However collision detection based on sensor data is com plicated by the sails and boat motion Particularly small boats cannot be equipped with sophisticated sensors e g due to weight and power limitations One approach to overcome this problem is to collect and store data from all participating vessels in a central data store This World Server then provides the data to all boats i e all participants in a race have access to a global view of the race situation We present our basic server implementation and first test results indicating that the approach allows implementing and testing collision avoidance without the need for bulky and expensive sensors 1 Motivation In a fleet race many sailboats compete with each other in a small area resulting in a high chance of collisions Human sailors have the ability of recognizing the positi
161. imilar conditions Red and green indicate values on starboard and port tack respectively 1 e with the apparent wind angle from 30 150 or 210 330 Black and blue indicate an apparent wind angle from 330 30 and 150 210 respectively www A TIBOOK ir 80 A Schlaefer et al that a careful calibration is required and also indicate that the quick motion of the boats needs to be considered when processing the compass readings 5 2 Performance In order to obtain data for different courses the boats were set up to sail in a rectan gle for which the longer side was aligned with the wind direction The rudder was controlled to maintain a constant target apparent wind angle which was changed in a round robin fashion First discrete wind angles with 10 degree increments were partitioned in windward and leeward courses Second two states were defined di rection and tack Whenever the boat crossed the windward side of the rectangle to wards the wind the direction was set to be leeward Likewise when the boat crossed the leeward side to the leeward the direction was set to be windward Similarly the tack was set to starboard tack when the boat moved to the right of the right side and to port tack when the boat moved to the left of the left side Third depending on the direction a new target apparent wind angle was selected from the respective partition every 20s To obtain a polar plot summarizing the speed with respect to the
162. implementation uses CHR rules for the following tasks e Expanding a node Every time a node is extracted from the open list a rule is triggering the creation of the neighboring nodes if they are not yet present e Creating edges If there are two neighboring nodes a rule creates an edge be tween the two nodes with the according travel time e Labeling neighbor nodes As soon as edges are present a rule labels the neigh bors of the node expanded last with the according cost and inserts them into the open list e Adding node to closed list If there are no more neighbors to be labeled a node is added to the closed list e Removing edges A node in the closed list triggers a rule removing all the incom ing and outgoing edges of this node e Path reconstruction Once the goal node is reached a CHR rule reconstructs the shortest path found We chose CHR for its declarativity which allowed us to implement the routing algorithm in a compact and clear fashion The implementation in CHR consists of only 17 rules with a little under 1000 lines of code In addition CHR allows the routing graph to be constructed and deconstructed dynamically by simply stating the conditions under which nodes and edges are created or removed as rules We furthermore believe that the implementation in CHR facilitates future adaption and extensions in an easy way which will help to incorporate changes that might be necessary on the ground of evaluations to come
163. imply tensioning and releasing control lines I contrast a wing sail is a rigid surface with an aerofoil cross section similar to an aircraft wing It can provide a much better lift to drag ratio than conventional sails 43 Neal et al state as significant disadvantages of a wing sail that it is extremely difficult to design it in a way that it can be reefed reliably Furthermore to construct strong lightweight rotatable wings at reasonable costs is mentioned there to be difficult However they maintain after extensive testing with different wing sails that the potential gains in reliability and efficiency outweigh these problems Although most of autonomous sailing boats featuring wing sails have been either designed for longevity or precision sailing 22 rather than for performance the America s Cup 2010 has shown impressively the dynamic abilities of a rigid wing sail The trimaran USA 17 formerly known as BMW Oracle Racing 90 or BOR90 won the trophy with a rigid wing as its main sail On a conventional sloop rig which is the most common rig type on sailing ves sels relatively high power is needed to tighten the sails against wind force As being self sufficient in terms of energy is one of the major goals in robotic sailing the rig design has been put into the focus of attention A balanced rig design also known as Balestron rig Aerorig swing rig and EasyRig provides great potential to save power 7 A balanced rig consi
164. in real life settings and conditions Our implementation can be used from the Prolog command line or via a Java application providing a graphical user interface The GUI visualizes the wind con ditions and allows to define starting and destination points on a world map It also provides configuration options for the routing and displays the calculated route on the map Figure B shows the GUI application www A TIBOOK ir A Rule Based Approach to Long Term Routing for Autonomous Sailboats 201 3 1 Evaluation and Related Work There are various commercial applications for long term weather routing of sail boats like the BonVoyage System MaxSea Sailplanner 9 or SailFast 8 The latter two are available as demo version thus we picked them to compare our rout ing algorithm to Both applications offer a GUI similar to ours While Sailplanner automatically downloads its own wind data and is restricted to one provider Weath erTech SailFast allows the usage of arbitrary GRIB files Sailplanner offers the user to pick from five different resolutions for the routing graph it uses while SailFast uses an isochron method with an unknown resolution but with configurable time steps However they are both closed source applications not revealing details about the algorithm they use for the routing We picked two routes of equal distance between starting point and destination about 1440 km on a great circle path one along the east coast of the U
165. ind since first of all there are no human on board which means less constraints with regards to security ergonomy or comfort In this paper two innovations are presented The first one addresses the main issue when navigating for a long time with a fully actuated sailing robot which is the energy used for guidance navigation and control One can use big solar panels to overcome that problem Instead we propose to use a wind vane self steering device Our idea is to put this device on the bow which simplifies its design This device regulates the trajectory of the boat relatively to wind direction As a consequence the wind sensor is not necessary The second innovation is a strategy of navigation between waypoints This paper will also present the prototype we used for testing those algorithms 1 Introduction Since many years unmanned surface vehicles USVs has been used for applica tions such as oceanography harbour protection or military as weapons or moving targets 2 USVs generally use propellers to move an get their energy from fossil fuels or solar power It is only recently that USVs use sail as a mean of propul sion 1 and also became autonomous The use of such vehicles is especially needed in oceanography 7 8 9 where there is a need for several months of data acqui sition in a designed area In fact the currently used buoys use to drift away very quickly from the data acquisition area and are often impossible to moor because
166. ing number of un manned system related projects hosted by the Systems Engineering department from underwater autonomous vehicles surface vessels and ground vehicles to a variety of aerial systems including helicopters planes and experimental craft The NavBoard3 uses a Rabbit 3000 processor core module programmed in a C programming design environment called Dynamic C The core module can be programmed with a special cable using a PC RS232 serial communication port or over a wireless interface if a special boot loader is used as is done with SailBot to provide in situ reprogramming The Rabbit 3000 processor although relatively slow when compared to many modern processors offers the advantages of a highly dependable In Circuit De bugger for development many built in software libraries and functions preemp tive and cooperative software multitasking capability large on board memory ca pacity of 512K bytes of Flash program memory and 512K bytes SRAM a Real Time Clock and a multitude of additional input and output capability A special NavBoard3 software library gives users access to the onboard hardware via stan dard C function calls allowing the user to focus on the higher level overall system implementation and control algorithms as opposed to the low level data acquisi tion and digital communication interfaces This control board encompasses a variety of sensors including a 3 axis acceler ometer and 3 angular rate gyros a 3 axis mag
167. ions This packet is received by Tracksail and used to set its actuator positions There is a slight loss of realism at this stage as this packet contains the target position of the actuator rather than its actual position This is due to the limitation that the Gumstix is not actually aware of the actual actuator position and can only send target positions to the PIC as part of a unidirectional data transfer This configuration was also constructed with the aim of accurately simulating and monitoring the power consumption of the robot To achieve this as with Bea gleB a set of hall effect current transducers are used to measure the current drawn by the actuators and other electronics These are connected to an Arduino microcon troller in the same manner as BeagleB except there is no current transducer from a solar panel Instead the solar panel current is calculated by Tracksail based upon the sun elevation for the current time date and location This is the represented as an 8 bit value 0 255 and is transmitted over the parallel port to a PIC18F4550 mi crocontroller The PIC converts it to a PWM signal which is in turn converted to an analogue voltage 0 5V using a filter circuit built from a resistor and capacitor This analogue line then connects to the Arduino replacing the input from the solar panel current transducer 4 Experiments and Results A series of experiments were carried out to test the difference between a pure Track sail
168. ir 148 D Wolter F Dylla and A Kreutzmann a b A 4x B c A 4x3 B head on A 4x40 B NA 4Z4 Ayind A B 4212 Byina Fig 7 Iconographic representation of the first steps in formalizing Rule 12 a 4 the vessels are in configuration r at the end 5 no collision occurs Other formalizations for example following shipping routes defined by buoys and light signals can also be formalized using the same approach Currently we do not regard disjunctive rules as for example in overtake situations where one is allowed to pass port or starboard The extension to include such formalizations is straight forward though 5 Navigation by Qualitative Rules In our approach qualitative representations serve exclusively for categorizing a con figuration and for checking rule consistency A randomized planner generates hy potheses of actions to perform only requiring the forward kinematics of the agent given by a sailing simulator in our case Actions generated by the planner are then assessed qualitatively actions that violate rules are discarded and the most promising actions are further considered Although a planner may be capable of de termining complex action sequences it is advisable only to exectute the first actions and to re plan as soon as possible Continuous re planning allows the system to re spond to unforeseen situations for example changing winds or unexpected behavior of others 5 1 Probabilistic
169. is paper There is one component however that is unique to USNA SailBot In the following section we will dis cuss that component and its implications for autonomous system control 2 2 The NavBoard3 As mentioned the form factor of most SailBot systems precludes the use of a lap top or netbook as the primary processor While we have used a small form com puter Pico system in the past the Technical Support Department TSD within the Systems Engineering department at USNA has developed a custom computa tion and sensing board that has a small form factor low overhead moderate cost and high capability The power and flexibility of this system has been proven in its use across a wide array of projects but none so thoroughly as the autonomous ro botic sailboat In this section we will discuss the key features of this crucial com ponent and provide the overall design concept behind its development so that the www A TIBOOK ir A Systems Engineering Approach to the Development of an Autonomous Sailing Vessel 97 interested reader can adopt and adapt the design principles to suit their needs and the wider needs of the autonomous vehicle community The Rabbit Navigation Board version 3 0 NavBoard3 is a programmable sin gle board computer that functions as an embedded controller for a variety of un manned systems requiring navigation based sensor feedback and output control signals The system was designed to facilitate the ever increas
170. itself This had the added advantage of allowing the rudder to be knocked out of alignment without risking any damage to the actuator Previous experience 9 in building sailing robots had highlighted the advantage of wing sails Wing sails provide a robust system that avoids the fragility of any lines connecting to the sail the potential of winches to slip and the need to either run lines through the deck or place actuators above the deck After the development of an initial prototype the potential for using these boats for researching control system design rig design rudderless sailing robots and au tonomous power management was realised This spawned a number of variations to the original MOOP design including boats with twin wing sails without rudders without wind sensors and a variety of wind sensor designs With the exception of the initial prototype MOOPO all other MOOP hulls have been built around a mould using glass fibre and epoxy resin A single piece perspex deck is screwed down onto the hull using 28 plastic screws A neoprene gasket helps to ensure that the join between the deck and the hull is waterproof A hatch in the deck built from a plumbing screw cap is used to access the power switch charging cable and data cables The mast connects to a Futaba S3306MG servo located inside the hull which is able to rotate it to adjust the sail position It passes through several oil seals to prevent water from entering the boat through
171. iversity of Liibeck Germany University of Applied Sciences Liibeck Germany University of Liibeck Germany University of Liibeck Germany University of L beck Germany University of L beck Germany Universidade do Porto Portugal University of L beck Germany Universidade do Porto Portugal University of L beck Germany ISEP IPP Portugal USNA United States of America University of Aberystwyth United Kingdom ETH Z rich Switzerland University of Aberystwyth United Kingdom INNOC Austria www ATIBOOR ir www ATIBOOR ir Contents Part I Introduction History and Recent Developments in Robotic Sailing Roland Stelzer Karim Jafarmadar Part II Robotic Sailboats Sailing without Wind Sensor and Other Hardware and Software TNO VATIONS 4 Ze ee ara are Sees ole Se A Jan Sliwka Jeremy Nicola Remi Coquelin Francois Becket de Megille Benoit Clement Luc Jaulin MOOP A Miniature Sailing Robot Platform Colin Sauze Mark Neal Breizh Spirit a Reliable Boat for Crossing the Atlantic Ocean Richard Leloup Frederic Le Pivert S bastien Thomas Gabriel Bouvart Nicolas Douale Henry De Malet Laurent Vienney Yvon Gallou Kostia Roncin A New Class for Robotic Sailing The Robotic Racing Micro Magic Alexander Schlaefer Daniel Beckmann Maximilian Heinig Ralf Bruder Part III System Development A Systems Engineering Approach to the Development of
172. kely hood of hurricanes by month 5 In July hurricanes are only likely to go as high as the Chesapeake Bay and hug the coastline However by August the storms are likely to extend up to Newfoundland and by September they can extend well to ward Greenland While various modifications to the basic great circle route were tried the best course from a climatology perspective will be the great circle route The path will pass well above the Azores high and the weather should be north of the high s di rect influence However if the high moves farther north the route could have less wind and possible regions of no wind for days 1 From the pilot chart the wind should be at a Force 4 for the majority of the transit with one region showing 31 of Force 5 conditions At the origin of the transit the wind direction should be from the southwest In the middle the winds should be westerlies and near the end of the crossing the winds should be from the west or northwest For most of the trip the vessel will be reaching or running Ireland has numerous potential harbors that would adequately serve as a finish location The four finalists were Fenit Valentia Kinsale and Castletown bere Fenit has good potential due to the proximity of the Fenit Harbour amp Marina to the ocean but the entrance is still protected It has little commercial traffic and the facilities and docks are more than sufficient for the re trieval The 1620 nm Great Circle Rout
173. l system for autonomous sailboats In IEEE International Conference on Fuzzy Systems 2007 www A TIBOOK ir Part III System Development www A TIBOOK ir www ATIBOOR ir A Systems Engineering Approach to the Development of an Autonomous Sailing Vessel Bradley E Bishop Joseph Bradshaw Cody Keef and Nicholas Taschner Abstract Over the past four years undergraduate students in the Systems Engi neering degree program at the United States Naval Academy USNA have pur sued autonomous sailboat systems development as part of their required capstone project in conjunction with students from the Naval Architecture program that de s gn and fabricate the vessels In this paper we discuss the pedagogy of robotic sailing as a capstone for Systems Engineering students and discuss our approach to interdisciplinary tasks such as this We also outline the design philosophy asso ciated with the systems side of the equation focusing on the on board instrumen tation and control hardware 1 The Pedagogy of Robotic Sailing Competitions The United States Naval Academy USNA is a baccalaureate institution that pro vides to every student both a degree and a commission in the United States mili tary primarily the Navy or Marine Corps Students in the Weapons and Systems Engineering department are required to complete a capstone project as part of their ABET accredited degree in Systems Engineering The students take a semester course on pr
174. lar results for scenarios 3 and 4 show that the influence of pruning away not allowed configurations does not interfere with action planning As the planner does not include any pre defined behavior e g how to sail against wind or how to start sailing a vessel from a complete standstill if facing the wind such basic sailing maneuvers have to be continuously re discovered by the plan ner This lack of expert knowledge can also explain the poor performance in sailing the narrow passage in scenario 2 While excluding such basic knowledge from the planning step may sound like an artificially created difficulty at first it significantly hints at the capability of the presented approach when moving closer to a realistic sailing simulation and ultimately when applying the method to a real autonomous sailing vessel With respect to a reasonable code of practice for basic sailing tasks we believe that the abstract qualitative representation provides solid means to for malize such general rules Our current approach can be extended to accommodate for such general rules The key difference between right of way rules and rules of good practice is that the latter kind only provides default knowledge that may be violated www A TIBOOK ir 154 D Wolter F Dylla and A Kreutzmann 7 Conclusion This paper demonstrates how navigation rules can be formalized with qualitative constraint calculi and how qualitative reasoning can contribute to solving t
175. lian endocrine system which is responsible for controlling a number of biological processes within the body These experiments showed that it was possi ble to use an artificial endocrine system to modulate the magnitude of rudder and sail actuator movements to control power consumption of a sailing robot and that these could be adjusted in response to internal conditions such as battery level or external conditions such as sunlight levels Further discussion of these results can be found in 12 These experiments also provide a large dataset on how well the MOOP robots perform In each experiment run the boats sailed a beam reach back and forth along a 250 m long roundtrip north south course for between two and four hours Across a total of 25 experiment runs 64 65 hours of sailing was under taken and a total distance of 61 27 km was covered This leads to an average speed of 0 95 km h The mean wind speed based upon only visual observations was 8 8 knots 16 km h with the minimum being approximately 3 5 knots 6 5 km h and the maximum being approximately 16 5 knots 30 km h Given the small size of the lake the maximum wave fetch can only be approximately 200 m and this tends to limit wave heights to below 10 cm but also causes very short wave periods typically less than 1 s It was originally intended to run these experiments by sailing triangu lar courses which would include both upwind and downwind sailing instead of just reaching Howeve
176. lity of success as either was feasible for a SailBot 2 The Vessel Inherent in a detailed analysis of either general route was a prediction of the boat s performance This is critical in two areas the upwind performance which will de termine whether the boat can sail against wind and current and the seaworthiness whether the vessel could survive expected storms Of the two the first is more dif ficult to achieve in a two meter boat The Spirit of Annapolis SOA a 2 meter long SailBot designed by the students was used in this project The boat has a displace ment of approximately 50 kg and a sail area of 2 4 square meters The size was chosen to fit within the SailBot Class Rules and ease logistics An earlier SailBot paper by Miller and his students 4 noted that SailBots are actually larger than the vessel which currently holds the smallest vessel to cross the Atlantic record SOA was designed primarily for seaworthiness rather than speed a departure from pre vious USNA SailBots To predict her speed a velocity prediction program PCSail was used VPPs use wind speed and direction as inputs and through parametric analysis of the boat s characteristics determine the forces generated A force and moment balance then yield the speed heel and other factors for the combination of windspeed and relative wind direction Using the Beaufort Force values for wind so as to correlate to the U S National Oceanographic and Atmospheric Administr
177. ller board and some other components 3 2 Sensors The boat gets data from three sensors GPS compass and wind direction simi lar to 1 The GPS is a small module from NAVILOCK Germany based on an ublox5 chip set u blox Switzerland Under free air conditions the module is able to generate differential GPS data Fig 4 b With a free sight to the south we try to fee S89 iia z SRE RE RRR RR REE TE B t e IE ETHERNET a cama 7 7 SS i rer at La H cee it a mi E rE oo we u i m P a im SA A kn aT oi y awd i an y tek lier pee u ze ml L Dem I Fig 4 Components a Controller board EK LM3S6965 with test adapter b GPS ce compass d Bluetooth modem www ATIBOOR ir 106 M Koch and W Petersen Fig 5 Rudder servo and sail winch in the test system Saudade receive the reports from the EGNOS satellite ARTEMIS to increase the accuracy of the GPS Unfortunately the connection to the EGNOS satellite is very unstable The GPS module occupies one UART interface As compass we use the 3 axis compass HMC6343 Honeywell USA It returns the tilt compensated value for the heading and the heeling of the boat It communicates over the I2C bus To get the direction of the apparent wind we use the PWM signal generated by the AS5040 austriamicrosystems Austria To eliminate the tolerance of the PWM frequency the controller gets the absolute value of the wind direction w_dir by the follo
178. lly the mentioned controller and operating system and the interaction between the components We show the system architecture of the control system the power consumption of the components and the different possibilities to process sensor data with the controller and the operating system 1 Introduction Working at an university located at the seaside and sailers since many years we are searching for a student project that combines practical experience in mechanical and electrical engineering and computer science with water and sailing In September 2010 we discussed the idea to build an autonomous sailboat with some students and now in the actual semester we have a small group of students who are building the boat The interdisciplinary team consists of four students of mechanical engineering and four students of electrical engineering and computer science The mechanical Michael Koch FH Stralsund Electrical Engineering and Computer Science Zur Schwedenschanze 15 18435 Stralsund Germany e mail michael koch fh stralsund de Wilhelm Petersen FH Stralsund Mechanical Engineering Zur Schwedenschanze 15 18435 Stralsund Germany e mail wilhelm petersen fh stralsund de www A TIBOOK ir 102 M Koch and W Petersen engineering group is building the boat and the mechanical parts while the sensors the controller and the software are built by the computer science group The stu dents are supported by the workshops of both de
179. mple and classical one by 9 For the forces on the rudders and the keel we use the model derived from air foil theory proposed by 10 The dimensioning case is a boat speed of 10 knots and the rudder moving alter natively from starboard to port with 20 degrees in magnitude The time to move the rudder from starboard to port is about one second The rudder law is driven accord ingly to the zigzag test ITTC standards Fig 3 shows that for such a configuration the force acting on the rudder reaches 200 N 2 3 Choosing the Rudder Motor After a study to determine the necessary torque for the rudder motor we selected the following actuators e servo motor Hitec HS 805 BB e servo motor Dynamixel RX 28 e servo motor Futaba Robbe BLS 151 www A TIBOOK ir 60 R Leloup et al Table 2 Comparison Table Servo motor model HS 805 BB Hitec RX 28 Dynamixel BLS 151 Futaba Criteria Robbe durability More than 100 hours More than 100 hours More than 1000 hours Waterproofness Designed for wet en Designed for dry en Designed for wet en vironment vironment vironment Nominal torque 19 8 24 7 kg cm 28 3 33 7 kg cm 9 6 kg cm Maximal torque 79 8 kg cm 113 2kg cm 14 4kg cm Input voltage 4 8 6V 12 16V 4 8V The comparison table shows various criteria including e Durability which is an important factor when considering the crossing of the Atlantic Longevity depends on the technology used Indeed brushless motors like RX 28 Dynamixel hav
180. mpliant Navigation with Qualitative Spatial Reasoning 151 1 iia d viwa Vfwa h Cfr Cfr Chuti m S friction d CupCsail ws CdriftCsail wl h Cfr Chull cr a buoyant drift drot Crot Vrot Vrot F cr h Viw sin r CdriftCsail w w s m a Nn Nm mm friction rudder wind rotation We have determined all parameters empirically by selecting values that yield reasonable sailing behavior the values are listed in Fig 8 While the model can easily be extended to include effects like currents changing winds etc it is suffi ciently complex to give the appearance of sailing as well as to require sophisticated planning techniques We note that the simulation needs to be realized as a efficient constant time step simulation since the roadmap planner needs to compare actions om a variety of different situations 6 Experimental Evaluation From the great variety of possible planning tasks we selected some interesting sce narios For each scenario we randomly instantiate planning tasks by varying the global wind and the speed of additional vessels involved in the scenarios We keep the courses of additional vessels fixed to gain independency of multi agent aspects We have selected the following types of scenarios see Fig 9 for illustration 1 Sailing along a straight route Sail from 0 0 to 100 100 on a 100m wide route with wind from an arbitrary direction 2 Sailing along a narrow
181. mputational Intelligence De Montfort University Leicester LE 9BH United Kingdom e mail rstelzer dmu ac uk Karim Jafarmadar INNOC Austrian Society for Innovative Computer Sciences Haussteinstra e 4 2 1020 Vienna Austria e mail karim jafarmadar innoc at www A TIBOOK ir A R Stelzer and K Jafarmadar by calculating an optimum route based on weather data and going on to independent tacking and jibing and avoiding collisions stand alone sailing boats are able to sail safely and reliably through to any and every destination The human being merely has to enter the destination co ordinates The key characteristics of a robotic sailing boat can be summarized as follows Wind is the only source of propulsion It is not remotely controlled the entire control system is on board It is completely energy self sufficient this is not a must in the sense of definition of a robotic sailing boat but t opens a wide range of applications Although many technical aids are available for common sailing boats relatively little effort has been spent on research of autonomous sailing Research on au tonomous surface vehicles ASV was mainly focused on short range crafts powered by electrical or combustion engines Such crafts are limited in range and endurance depending on the amount of fuel or battery capacity on board to run a motor for propulsion In contrast a sailing vessel needs only a minimal amount of power to run sensors compu
182. ms 2007 www ATIBOOR ir Simulating Sailing Robots Colin Sauz and Mark Neal Abstract This paper outlines our experiences in simulating sailing robots It fo cuses on efforts to simulate a sailing robot both in pure software and a Hardware in the Loop HIL simulation and compares these results with sailing a similar course using an actual robot The software simulator is built upon an open source sailing game called Tracksail This is based on a highly simplistic physics model and makes no attempt to simulate the actions of waves currents or tides The target robot to be simulated is BeagleB a 3 65 m long boat based upon a MiniJ dinghy equipped with a 2m high wing sail To provide a more realistic simulation of this boat a Hardware in the Loop simulator was constructed using nearly identical electronics attached to a trolley table Sensor inputs can either be provided by a Tracksail simulation or from sensors on the table Sailing of an identical course in the Tracksail the HIL System and on BeagleB showed similar results in all cases This suggests these simulations are a reasonable approximation of real world robot performance Although they are no substitute for actually sailing a robot they are a sufficient for testing software testing hardware and developing control strategies 1 Introduction Simulators offer an opportunity to develop and test sailing robot control algorithms without incurring the overheads of deplo
183. ms and Sensors Workshop 2003 http www geo prose com ALPS white_papers alt pd www A TIBOOK ir 20 ioe O 11 12 13 14 15 16 17 18 19 20 R Stelzer and K Jafarmadar Alves J Ramos T Cruz N A reconfigurable computing system for an au tonomous sailboat In IRSC 2008 International Robotic Sailing Conference pp 13 20 2008 http www roboticsailing org fileadmin user_upload _ temp_ Roboticsailing Proceedings web pdf page 13 Ammann N Hartmann F Jauer P Bruder R Schlaefer A Design of a robotic sailboat for wrsc sailbot In International Robotic Sailing Conference 2010 pp 40 42 2010 BalancedRig Balancedrig website 2009 http balancedrig com description html accessed April 16 2009 Benatar N Qadir O Owen J Baxter P Neal M P Controller as an Expert System for Manoeuvring Rudderless Sail Boats In Proceedings of UKCI 2009 http www users york ac uk oq500 pdfs UKCIO9 paper pdf Bennett S A history of control engineering 1800 1930 Inspec lee 1986 Bertram V Unmanned surface vehicles a survey Skibsteknisk Selskab Copen hagen Denmark 2008 http www skibstekniskselskab dk public dokumenter Skibsteknisk Download20materiale 2008 1020marts2008 USVsurvey_DTU pdf Blidberg D The development of autonomous underwater vehicles auvs a brief sum mary In IEEE ICRA vol 4 Citeseer 2001 http citeseerx ist psu
184. n in a competition can be both a blessing and a curse While the students are highly motivated to win it is easy for them to lose sight of good design principles and the requirements of a capstone project in the face of the immediate demands of the competition It is our belief that winning the com petition should never be the primary goal but rather should be a consequence of a proper design experience that begins with an appropriate needs analysis While it seems logical that the winning entry in a competition will of needs be the best designed of the field it is not necessarily the case Stop gap measures kludges and inelegant coding wiring and design are the hallmarks of projects whose only metrics are victory at any cost While one of these vessels may win a competition that spans a few days they will often have deep design flaws that may not be obvious from their performance These flaws in design are the hallmarks of a failed capstone project one in which the students did not learn the appropriate lessons of design and built an overly inflated view of kit bashing and caffeine fueled desperation that will not stand the test of a real world engineering problem It is therefore essential that students engaged in a capstone design whose output is to be used in a competition be guided appropriately through the proper engi neering design and be evaluated on the same This is a problem of perception on the part of the students and expectation o
185. n the part of the advisor s and the host department both of which need to be managed carefully from the outset of the project 1 2 Project Execution Methodology SailBot SailBot is a student driven competition involving automated sailboats that com pete in racing navigation and station keeping contests SailBot class vessels are 2 m or less in length while Open Class vessels can be up to 4 m in length Full de tails of vessel requirements can be found in 2 It is important to note that the SailBot project at USNA is an interdisciplinary effort between the Systems Engineering department and students pursuing a Naval Architecture degree The Naval Architecture students design and build the vessel in a two course sequence Fall Spring and the Systems Engineers provide in strumentation actuation and automation as a follow on effort the next academic year This process has yielded quality results for many years now and there are important lessons that have been learned In this section we will discuss the for mal pedagogy of SailBot as a capstone design project for the Systems Engineering students starting with a completed but strictly R C vessel generated by the Naval Architecture students www A TIBOOK ir A Systems Engineering Approach to the Development of an Autonomous Sailing Vessel 89 Students in Systems Engineering at USNA are expected to formally define a full set of objectives metrics functions and demonstration plans for
186. n with the Gumstix and wireless network equipment using between 400 mA and 800 mA at 5 V This accounts for a significant proportion ofthe robot s power consumption and restricted run times to approximately 4 hours using 18 2700 mAh batteries connected in 3 parallel packs During summer 2009 MOOPO and MOOPI underwent a significant amount of testing to improve their control systems In this process several hardware changes were also made The rudders of both boats were doubled in area this helped the boats to turn at low speed usually after a failed attempt to tack or in strong winds Two LEDs were placed inside the hull immediately underneath the deck one at the stern and one at the bow These helped to locate and determine the direction the boat was facing during darkness or heavy fog The wind sensors were changed from potentiometers to more accurate rotary magnetic encoders 2 2 MOOP2 MOOP 2 was constructed in summer 2009 with the aim of sailing across the At lantic Ocean in the later cancelled 2009 Microtransat Challenge To meet the power budget requirements it was equipped with 10 size F 1 25 V 13 Ah bat teries connected as 2 parallel packs of 5 batteries Solar cells were attached to the underside of the front half of the deck these were expected to supply an average of only 80 mW of power The transatlantic journey was expected to take approximately 200 days and assuming the batteries started the voyage fully charged and ended it
187. nd future role in sustained ocean observations Marine Technology Society Journal 43 1 21 30 2009 Sauz C Neal M Design considerations for sailing robots performing long term autonomous oceanography In Proceedings of The International Robotic Sailing Con ference May 23 24 pp 21 29 2008 http www innoc at fileadmin user_upload _temp_ Roboticsailing Proceedings web pdf Scanmar Classification of Vane Gears by Course Correcting System 2011 http www selfsteer com windvanes101 classification php accessed on April 27 2011 Schieben E W Skamp an amazing unmanned sailboat Ocean Industry pp 38 43 1969 Shukla P Ghosh K Revival of the Modern Wing Sails for the Propulsion of Commercial Ships International Journal of Civil and Environmental Engi neering 75 80 2009 http www akademik unsri ac id download journal files waset v1 2 14 19 pdf Sliwka J Reilhac P Leloup R Crepier P Malet H Sittaramane P Bars F Roncin K Aizier B Jaulin L Autonomous robotic boat of ensieta In 2nd Inter national Robotic Sailing Conference Matosinhos Portugal 2009 http media ensta bretagne fr robotics images 4 4f Ensieta_team pdf Smith B Skamp roboat boat with rigid sails patrols ocean beat Popu lar Science 196 5 70 72 1970 http books google at books 1d 8QAAAAAAMBAU amp pg PA7 0 amp dq SKAMP Schlieben amp source gbs_ toc_r amp cad 2 v onepage amp gq SKAMP20Schlieben amp f f
188. nd of the message 2 sent by a client to the World Server to request the server to send all registered objects cf 1 above 3 currently not used 4 sent by a client to the World Server followed by an object to register update at the server 5 sent by a client to the World Server to remove an object from the server currently not allowed Each object is encoded in a single line ended by n or n r respectively follow ing a simple scheme lt timestamp gt TYPE lt object type gt ID lt identifier gt Lat lt latitude gt Lon lt longitude gt lt type specific fields gt n Timestamp Type ID Lat and Lon are mandatory for every object but all type specific fields are optional and will be set to 0 by the World Server if not given Parsing objects can be done easily by handling each line separately and first splitting the string at each and then at lt gi The following C code the Java code looks nearly the same demonstrates the easy usage of our client It connects to the World Server sends a SailBoat object to the server then gets all objects from the server and disconnects Both options are allowed 2 The difference measured in milliseconds between the current time and midnight January 1 1970 UTC 3 Look at our source code available on www wrsc2011 org to see how this is done www A TIBOOK ir 164 N Ammann et al Create a new WorldServerClient to connect to the World Server running on
189. ndependent control strategy for au tonomous sailboats based on voronoi diagram 2009 www ATIBOOR ir www ATIBOOR ir Breizh Spirit a Reliable Boat for Crossing the Atlantic Ocean Richard Leloup Fr d ric Le Pivert S bastien Thomas Gabriel Bouvart Nicolas Douale Henry De Malet Laurent Vienney Yvon Gallou and Kostia Roncin Abstract To meet the Microtransat challenge ENSTA Bretagne chose to realize several sailing robots The first having served as a test platform allowed us to de velop two new boats one for research and one for the Atlantic crossing The dif ferent tests in the bay of Brest allowed us to improve the reliability of systems on our sailboat Various studies have been conducted to improve reliability of sailboats both mechanically and electronically This tests allowed us to test different concepts on the three Breizh Spirit boats The results are very positive and we can now say that we have a boat able to resist to strong storms to follow a predifined route to supply its own energy and to navigate in sea waves From the experience acquired from Breizh Spirit 1 we hope we will be able to cross the Atlantic Ocean 1 Introduction As part of the international Microtransat challenge between different universities and scientific schools ENSTA Bretagne decided to develop several sailing robots In 2009 ENSTA Bretagne developped a 1 3m sail boat Fig 1 named Breizh Spirit 1 The boat meets several of the re
190. ndomized qualitative approach demonstrat ing that reasonable control commands can be determined to control an autonomous robotic vessel in a rule compliant manner In future work we aim to reproduce our results in a sophisticated simulation con text stepping closer to control a real autonomous vessel We plan to extend the qual itative rule formalization by high level description of navigation recommendations to improve sailing performance see 16 While we currently use a simple model to anticipate the actions of other agents interesting scenarios like regatta racing call for a much more involved handling of multi agent aspects We are confident that the qualitative rule formalization provides excellent grounds to tackle such competitive multi agent navigation problems Acknowledgements This work is supported by the Deutsche Forschungsgemeinschaft DFG in context of the transregional collaborative research center SFB TR 8 Spatial Cognition project R3 Q Shape Financial support is gratefully acknowledged We also acknowledge the comments of the anonymous reviewers References 1 Belghith K Kabanza F Hartman L Using a randomized path planner to generate 3D task demonstrations of robot operations In International Conference on Autonomous and Intelligent Systems AIS pp 1 6 2010 2 Belghith K Kabanza F Hartmann L Nikambou R Anytime dynamic path planning with flexible probabilistic roadmaps In Proceedings IE
191. netic field sensing module for deriv ing heading an on board Xbee wireless communication transceiver four servo control ports with programmable pulse width modulation and a GPS receiver These sensors provide the ability to collect navigation information and sense exter nal body rates and forces for robotic platform state estimation There are many un dedicated I O ports to which the user can interface external sensors both analog and digital The NavBoard3 also has a variety of external communication bus inter faces such as I2C SPI and serial UARTs making it easy to interface with almost any COTS product The overall layout of the custom board is shown in Fig 1 The NavBoard3 is modular and easy to maintain with most components being easily removable for replacement The components are all themselves COTS and the PCB is manufactured in bulk This device uses the processor that is the core programming and control learning platform for the department reducing the learn ing curve for students adopting it as their processor The integrated IMU capabili ties reduce overall system complexity for autonomous vehicles and allow for higher level work on the part of the students At the same time any and all capa bilities embedded in the board can be supplanted by external devices as needed www A TIBOOK ir 98 B E Bishop et al Rabbit Navigation Board 3 0 System Layout Power 3 Axes On Board 3 Axis LED GPS RTC Input ADXRS610 Linear Power
192. nics has been developed based on PIC microprocessor PIC 18F2550 to have enough memory and calculation power The sensors required to control the boat are chosen for their performance and low consumption e Wind sensor Ultrasonic anemometer CV7 LCF Capteurs FRANCE e GPS FV M8 SANAV San Jose Technology Inc USA e Magnetic Compass HMC6343 Honeywell International USA The servo motors Fig 2 are powered by a circuit different from the micropro cessor on the centre of Fig 6 The energy supply can be put on stand by by the microprocessor n order to limit losses load current supply and servo motors for www ATIBOOR ir Breizh Spirit a Reliable Boat for Crossing the Atlantic Ocean 63 Fig 6 naBreizh Spirit 3az s new electronics the moments All power supplies loaded conversion of 12 V 5V 3 3V etc are selected for their efficiency 0 93 This energy management can lower the aver age consumption of between 100 and 150 mA The embedded energy is stored in a battery of 12 Volts 12 Ah on the right of Fig 6 Conventional lead acid battery technology was chosen because of its reliability Such a battery without electrolyte fluid is sufficient for a journey expected to last six months This system can supply the energy for 3 days minimum The recharge of the battery is provided by solar panels that can charge up to 500 mA These panels can charge the battery during the night operation with a slight margin for the days o
193. nloading of the latest GRIB files 6 Conclusion Preparing for a successful autonomous crossing of the Atlantic requires more than designing and building a good platform A higher probability of success can be as sured by choosing a route that minimizes potential weather problems Two potential routes each identifying a relatively high degree of success for small autonomous vessels were identified Further research is needed to identify the specific destina tion for the Northern Route References 1 Cornell J World Cruising Routes International Marine Pub Camden 1987 2 Defense Mapping Agency Atlas of pilot charts North Atlantic Ocean Prepared from data furnished by the Defense Mapping Agency of the Department of Defense and by the Department of Commerce 1994 3 Fisheries and Oceans Canada The Grand Banks of Newfoundland Atlas of Human Ac tivities Commercial Shipping Traffic Density 2000 cited June 3 2011 4 Miller P Brooks O Hamlet M Development of the USNA SailBots ASV In 2nd International Robotic Sailing Conference Porto Portugal 2009 5 National Oceanic and Atmospheric Administration s National Weather Service 2010 Tropical Cyclone Climatology http www nhc noaa gov pastprofile shtml cited April 7 2011 www A TIBOOK ir A Rule Based Approach to Long Term Routing for Autonomous Sailboats Johannes Langbein Roland Stelzer and Thom Fr hwirth Abstract We present an algor thm fo
194. ns students skip over the vital step of defining design objectives The task presented to the students is to de fine the objectives and then push toward the functions that support those objec tives The following is a discussion of the primary objectives that are crucial to a good SailBot system design in bold with the high level functions that are needed to support them There are some additional design objectives and further refinements that are not listed here due to space restrictions The system must be 1 Competitive this objective focuses on the tasks that are required for victory in the SailBot competition The vessel and its controller must be a Fast the system must achieve its navigation and sailing tasks quickly in order to compete b Accurate the system must minimize error in navigation and in executing selected sailing commands c Energy efficient the system must not waste battery power as some of the tasks are long duration 2 Reliable the systems must be designed to reduce maintenance and update costs in time and money a Robust the system must be capable of carrying out its mission with minimal sustained damage over many cycles Must be concerned with www A TIBOOK ir 90 B E Bishop et al temporary power loss water intrusion damage from slamming waves and possible collisions b User Friendly the system must admit quick setup configuration power recharge component replacement and in
195. ns the actuators The Gumstix generates UDP telemetry packets which include actuator position data This is fed back to Tracksail running on the laptop so that it can change its actuator positions The Gumstix also receives data on power consumption from the Arduino which processes data from the current trans ducers As there is no real solar panel on this system its current transducer output is simulated by a PIC controlled via the parallel port of the laptop www ATIBOOR ir 120 C Sauz and M Neal Figure 3 shows the data flows for the process Tracksail generates a UDP packet every time it completes its main loop which contains the simulated boat s heading and position and the wind direction These are then received by another process running on the laptop which applies noise to them and generates data in the NMEA 0183 format used by the real sensors The noise is created using a Gaussian distri bution with wind and compass data being subjected to noise ranging between 2 degrees and GPS data ranging between 5 metres To match the real sensors GPS and wind data are limited to transmission at Ihz and 4800 bits per second and compass data at 4hz and 19200 bits per second This sensor data is then received by the Gumstix and processed by the same code as run s on BeagleB To complete the loop the Gumstix then transmits a UDP packet normally used to send telemetry data back to a monitoring station which contains the actuator posit
196. ns were printed on paper from a design drawing and cut into shape These were then glued to sections of polystyrene approximately 5 cm _ Wind Vane sie ee Wind Vane 13cm LE 12cm Fig 1 A diagram showing the dimensions of a MOOP hull and sail www A TIBOOK ir 42 C Sauz and M Neal thick and cut using a hot wire cutter to match the shape of the cross section printout Each of these sections were then joined together to form a polystyrene hull this was then smoothed off and covered in glass fibre and epoxy resin The inside of the hull was then hollowed out where required This resulted in much of the hull still being full of polystyrene this added buoyancy but restricted the amount of space available inside the hull The use of relatively coarse polystyrene also left a rough feeling to the entire hull As shown by the photograph in Figure 3 the deck was split into two sections with a section of glass fibre across the middle between them The mast was placed in the centre of the boat between the two deck sections By placing the mast in this location it was hoped that it would reinforce the mast An aluminium bulk head was added beneath this middle section to split the inside of the boat into two compartments Two holes were made near the top of this bulkhead to allow cabling through but it was hoped that this would prevent water from spreading through the boat in the event of the hull being breached or leaks occurring thr
197. nt This drift is due to a bad setting of the rudder servo This servo was changed for a better model just before the test and we just forgot to do new controls As a consequence for each way point the boat had to tack and go upwind to reach its way point The validation circle is about 400m diameter When it was about to reach the Swensea Vale Way point the wind suddenly strengthened up to 30 knots As we can see the boat was no longer able to tack into the wind Then it was tacking upwind for approximately 10 hours before it crashed on the rocks Due to problems of accessibility the boat was left there for three days Two storms destroyed the entire rig and appendages but the electronic system was intact despite some water ingression 6 Conclusion As a conclusion we can say that the experience gained from Breizh Spirit 1 has enabled us to design a boat Breizh Spirit 3 much more suited to cross the Atlantic Indeed we found some errors in the design of the Breizh Spirit 1 The electronics has Table 3 Progress Table of Breizh Spirit Test Platform BS1 Validated Electronic architecture for BS1 and BS3 Validated Autonomous navigation of 12 miles with BS1 Validated Energetical design for 6 months of BS3 Validated Mechanical design for Crossing the Atlantic of Validated BS1 and BS3 Control of component s reliability with BS1 Validated Setting Controls for BS3 In progress Endurance test with BS3 In progress Construction of BS3 In progres
198. ny navigable weather condition 5 User Friendly again this metric scores on a sequence of one point binary de cisions for a total of 0 4 points 1 allows wireless programming and reconfiguration 1 allows for battery replacement underway www A TIBOOK ir 92 B E Bishop et al 1 allows for one point disconnect of any device or subsystem 1 allows for one point access to any subsystem or component It is well and good to define a set of metrics but it is necessary that the system be designed to meet those metrics and that there be a plan for testing the perform ance We utilize the pairwise comparison chart PCC morphological chart and decision matrices to select components to achieve the functions and hopefully meet the metrics 1 Briefly the PCC provides a weighting for each objective relative to each other objective The morphological chart enumerates possible me thods by which each function can be achieved e g sensing could be a GPS and an IMU or it could be a weather station there might be sensors for sail pressure or for relative wind speed etc A decision matrix is generated for each subsystem by taking each potential implementation for that subsystem and scoring it based on the appropriate metrics An overall score for each possible solution is generated based on the scores and on the appropriate objective weightings The potential so lution with the highest composite score for each subsystem is the prelimin
199. o cross the wind line We are currently working on strategies to force the tacking meneuver when the speed of the boat is null The idea is to block the rudder and use the wind vane to push the boat backwards If the rudder is blocked in a suitable position the tacking maneuver will be accomplished www A TIBOOK ir 32 J Sliwka et al a b 7 gt een nn oath Page i i a ae gt 4 aa a i Fi 1 Fa EN Ya i XY U y pa A pa EA i ni I i Fig 4 Executing jibbing and tacking maneuvers using the self steering device SCILAB simulation Sub Figure a corresponds to the case where the self steering device is on the bow Sub Figure b corresponds to the case when it is on the stern 2 3 Mechanical Design In this part we explain in more details the mechanical design of the bow wind vane self steering system Installing the system on the bow simplifies the design of the system since there is no need for gears to invert the rotation of the wind vane and the rudder As for the robustness installing a mechanism bow put it under the action of the wave which might destroy it However in case of a bow self steering device the most fragile part which is the wind vane will align itself with the wind and the waves reducing their impact on the structure Figure 5 represents the overall design of the mechanism The mechanics is really simple However because the robot is small the linkage has to have less friction than
200. o much on the weather conditions We found the Graupner Micro Magic kit to be a good compromise with respect to the aforementioned criteria The boat is small inexpensive and widely available Having built a fleet of five boats we feel confident that the Micro Magic presents a good basis as an entry level robotic sailing class One of our modified rrMM boats is shown in Figure l While a strict one design would be ideal to compare the performance of the al gorithms boats need to be adapted to different conditions to sail efficiently e g by weight trim or reefing We therefore deliberately allow some degrees of freedom in building and trimming an rrMM class boat and propose the following rules as a basis for competitions www A TIBOOK ir A New Class for Robotic Sailing The Robotic Racing Micro Magic 73 Fig 1 The figure shows a lateral and top view left and center of one of our boats On the right hand side a more detailed view of the deck layout and the electronics is shown In contrast to the conventional design we have mounted the servos on deck and our control board to the lid Both servos are sealed and a carbon fiber rod is used to connect them to the boom fittings The black box inside the hull houses the GPS unit while the black part in the cockpit is the GPS antenna 1 The hull and deck must be from the Graupner kits either MicroMagic or racing MicroMagic 2 The hull must be unmodified while the deck layout may be c
201. o the current passing through them these voltages are processed by an Arduino Uno ATMega328 microcontroller which is able to pass this data on via an ethernet interface to the Gumstix in the form of a UDP packet sent every 10 seconds This process allows for power consumption to be tracked over time 3 1 2 Long Term Power Management BeagleB is intended for the dual purpose of performing autonomous oceanographic monitoring and researching long term power management in sailing robots Sail ing robots offer a unique opportunity to for relatively low cost and low risk research power management strategies for robots which must operate without contact and su pervision from human operators for prolonged periods of time The relatively small size of BeagleB and the general desire to minimise size cost and potential danger to other vessels limits the amount of power which can be generated by onboard Table 1 The specifications of BeagleB Name BeagleB Hull 3 6 5m long Mini J dinghy Keel 80 kg single keel Sail 2 m high carbon fibre wing sail Sail Actuator LINAK LA 12 Rudder Actuator LINAK LA 12 Computers PIC18LF4550 and Gumstix Wind Sensor Furuno Rowind Batteries 2 8 kWh sealed lead acid batteries Solar Panels 2 panels 90 W peak total Compass Furuno PG 500 Fluxgate Compass GPS Furuno GP 320B Power Tracking 4 LEM CAS 6 NP Transducers sampled by an AT Mega328 Microcontroller www A TIBOOK ir 116 C Sauz and M Neal photovolta
202. odules will be included in future 4 Software The software of the control system is written in C TI provides a driver library to encapsulate some of the hardware details e g to program a timer or to commu nicate using an UART To simplify the software development there are different programming environments available In this project we use the IAR Embedded 3 IAR Systems Sweden www ATIBOOR ir 108 M Koch and W Petersen Workbench for ARM An alternative are TIs Code Composer Studio development tools Based on the in circuit debugging interface of the evaluation kit we are able to program and debug the controller from within the Embedded Workbench using the USB interface 4 1 Operating System Because of the different tasks the control system has to do during sailing get the sensor data calculate the course generate the signals for actuators and communicate with the environment we decided to use an operating system for the control system There are different operating systems available for the ARM7 controller We use the very small and stable operating system uC OS II It has a small real time kernel and has been certified within some commercial products to meet the requirements to be used in safety critical systems Details of the OS can be found in 2 An alternative with compatibility to Linux is RoweBots Unison V5 RTOS RoweBots Canada 4 1 1 Hardware Requirements To use uC OS II as OS for a microcontroller the
203. of ocean s depth Research in autonomous sailing boats was also stimulated by the Jan Sliwka Jeremy Nicola Remi Coquelin Francois Becket de Megille Benoit Clement Luc Jaulin ENSTA Bretagne 2 rue Francois Verny 29200 Brest France e mail jan sliwka jeremy nicola ensta bretagne fr www A TIBOOK ir 28 J Sliwka et al MicroTransat challenge 3 which purpose is to cross the Atlantic ocean in full au tonomy Many autonomous sailing robots builders tend to make copies of existing sailboats for mankind installing the same set of sensors and replacing human ac tuation by electronic actuators ENSTA Bretagne robotic team has chosen to work on sailing robots capable of navigation for long periods In order to reduce the en ergy consumption and to improve the robustness of the robot we focus on reducing the number of sensors and actuators especially the wind sensor and develop new algorithms and or mechanical designs to deal with that handicap In this paper we propose a solution for a robotic sailboat which addresses the main issue when nav igating for a long time with a fully actuated sailing robot which is the energy used for guidance navigation and control One can use big solar panels to overcome that problem Instead we propose the use of a wind vane self steering device This kind of device actually exists on real boats 4 but our approach is to install this device on the bow instead of putting it on the stern since it
204. of the team it uses a single traditional sail controlled by an electric winch system Pinta is based on a Toper Taz sailing dinghy with a length of 2 95 m The rudder is controlled by an off the shelf auto helm Pinta was the only boat to enter the 2010 Microtransat Challenge The MOOP Mini Ocean Observation Platform Figure is a small lightweight sailing robot with an overall length of 0 72 m Multiple MOOPs have been built so far with single or twin wing sail configuration with rudders and rud derless They are controlled either solely by a PIC microcontroller or with a com bination of PIC and Gumstix Single Board Computer Several MOOPs took part in the SailBot and WRSC competitions in 2009 and 2010 3 3 9 University of L beck University of L beck started their autonomous sailing activities in 2009 They par ticipated in Sailbot WRSC 2010 with two boats D umling Figure 6a is based on a University Club Graupner Germany boat model with a length of 0 53 m and two sails with a total sail area of 0 145 m The boat is inteded to be used as a testbed for various methods of artificial intelli gence Pi mal Daumen Figure 6p is based on a standard IOM International One Meter class hull with a length of 1 m and two sails with a total sail area of 0 4 m7 On board control is realised with a custom built circuit board featuring an AT Mega2560 microprocessor 6 3 3 10 University of Porto Team FASt Figure developed a 2 5 m lon
205. offer a means to debug and perfect software and hardware before they are deployed at sea in a more challenging environment where the cost of failure is far higher 3 2 Tracksail Al Simulator As discussed in section 2 in previous work we have modified the Tracksail open source sailing game to produce a sailing robot simulator This derivative known as Tracksail Al 3 allows a client program to connect to the simulator using a TCP IP socket It can then issue commands which allow it to control the rudder and sail po sitions read the wind direction boat heading and location The Tracksail Al server is written in Java and presents a graphical interface illustrating the wind direction and the boat s heading location rudder position and sail settings Additional client side code is used to calculate the state of a simulated solar panel based upon time of day time of year and location and to track the battery state based upon the move ment of actuators and which sensors are considered to be switched on up to date information will not be made available for any sensor which is considered to be off Through this power consumption data can be generated and experiments with artificial endocrine systems as discussed in the previous section undertaken In our latest version of Tracksail there has been a focus upon parallelisation with the abil ity to run multiple instances of the same experiment in parallel Jobs can also be distributed across multi
206. oject management in the Fall semester of their 1 C senior year In this course students propose a project and are assigned a faculty advisor In the following Spring semester students execute their plan and produce a fully integrated capstone project with a formal performance and design review Systems Engineering at USNA focuses on mechatronics robotics and feedback control As such our projects tend to be very hardware intensive and must include some form of autonomy or automation Due to the marine centric nature of our service many students opt to work on projects that involve autonomous marine vehicles both surface and underwater Many students choose autonomous sail boats for their capstone projects due to their familiarity with sailing or from a de sire to apply advanced control and automation in a competitive framework Bradley E Bishop Joseph Bradshaw Cody Keef Nicholas Taschner United States Naval Academy Annapolis MD USA e mail bishop bradshaw m116552 usna edu cody keef36 gmail com www A TIBOOK ir 88 B E Bishop et al 1 1 Robotic Sailing as a Capstone Project While autonomous sailing vessels have been developed at USNA prior to the ef fort discussed in this paper the origin of this work was initially the SailBot com petition 2 3 and later the World Robotic Sailing Championship WRSC 4 It has been the point of view the USNA Systems Engineering department for many years that student participatio
207. om 0x01 to 0x50 with Ox4D being the default value 6 2 4 Register 08 Interference Rejection Payload length 1 byte This option at register 0x08 accepts four values where O off 1 low 2 medium and 3 high 6 2 5 Register 0A Target Boost Payload length 1 byte The Target Boost setting at register OxOA accepts three values where O off l low and 2 high 6 2 6 Register QE Local Interference Filter Payload length 1 byte This option has four values where O off 1 low 2 medium and 3 high Table 3 Register values for zoom ranges Range Hex Data Range Hex Data 50m Oxf4 0x01 0 0 2km 0x20 0x4e 0 0 75m Oxee 0x02 0 0 3km 0x30 0x75 0 0 100 m Oxee 0x03 0 0 Akm 0x40 0x9c 0 0 250 m 0xc4 0x09 0 0 6km 0x60 0xea 0 0 500 m 0x88 0x13 0 0 8km 0x80 0x38 1 0 750m 0x4c 0x1d 0 0 12km Oxc0 0xd4 1 0 lkm _0x10 0x27 0 0 16km 0x00 0x71 2 0 1 5km 0x98 0x3a 0 0 24 km 0x80 0xa9 3 0 www ATIBOOR ir A Digital Interface for a Lowrance Broadband Radar 177 Table 4 Register values for filters and preprocessing Function Register Values 0 l 2 3 4 5 6 7 8 9 10 Automatic Gain 06 C1 00 00 00 00 01 00 00 00 Al reg cmd sel Manual Gain 06 Cl 00 00 00 00 00 00 00 00 xl reg cmd sel val Rain Clutter Filter 06 Cl 04 00 00 00 00 00 00 00 x2 reg cmd sel val Sea Clutter Harbour Automatic 06 Cl 02 00 00 00 01 00 00 00 D3 reg cmd sel Sea Clutter Manual 06 Cl 02 00 00 00 00 00 00 00 x3 reg cmd sel val 6 2 7 Register OF Scan Speed Payload len
208. om the eight major directions north northeast east etc averaged over more than one hundred years of data collection from ves sels Additional information includes current vectors and velocity that illustrate the prevailing currents present in that particular month The dotted current vectors in dicate variability especially found at the surface Sea ice is also shown providing information such as a maximum ice limit and the mean maximum iceberg limit The percentage of gales for every five degrees is shown which helps determine which regions will have a greater chance for Force 8 winds and above Once the shortest route was plotted on the pilot chart it was compared against other routes that might provide either shorter times or higher probability of suc cess For instance if an area on the great circle course showed a higher probability of headwinds or gales a route around that area would yield a higher probability of success The desired outcome that could be compared was the theoretical time the boat would take to cross the ocean The data required for this calculation in cluded the distance direction heading wind direction percentage and force and www A TIBOOK ir Route Planning for Micro transat Voyage 187 Table 2 Calculation results for Cell 1 of the shortest northern route VMG in this table can be compared to Table 1 noting that the vessel is not designed to sail closer to the wind than 46 degrees Section 1 Dist
209. on of each boat Furthermore they can predict the future movements by observing the surroundings so that they can react and avoid collisions In robotic sailing however observing the surroundings is a quite challenging task In spite of limited sensing abilities computing time and power supply on a robotic sailboat one has to find a way to determine where objects are e g obstacles or buoys Competitions in robotic sailing already exist 1 2 3 but up to now these have not included automatic collision avoidance So far collision avoidance in robotic sailing is done manually although some ap proaches towards active object detection on vessels have been introduced 8 9 In contrast to large commercial vessels where radar is a common navigation Nikolaus Ammann Florian Hartmann Philipp Jauer Julia Kr ger Tobias Meyer Ralf Bruder Alexander Schlaefer Institute for Robotics and Cognitive Systems University of Liibeck e mail sai ling rob uni luebeck de www A TIBOOK ir 158 N Ammann et al technology the installation of radar systems on small boats is quite challenging Another approach for object detection is to use cameras Our motivation is to develop a system that enables autonomous sailboats to be aware of their vicinity without the use of active monitoring sensors as those men tioned above To achieve this task the boats need information such as the position and heading of other boats nearby race course and race
210. on can be made a replacement for one or more of the incompatible systems is selected starting with compatibility with the remaining system as a prerequisite The process s carried out iteratively until a complete and compatible preliminary solution is obtained Finally students must generate a demonstration plan by which they will test each component of the system and measure the actual performance Again guid ance by the advisor is crucial during this step as students will tend toward the build the system and test the whole thing approach Students who work on Sail Bot are required to carry out the following tests on every subsystem www A TIBOOK ir A Systems Engineering Approach to the Development of an Autonomous Sailing Vessel 93 1 Component bench test Using a stable wall power supply and the least complex interface available manufacturer provided testing software simple RS232 etc demonstrate that the performance of the component meets its specs This will often involve generation of test inputs and parameterized test runs and will often result in extensive troubleshooting and sometimes a re turn to the decision matrix for an alternative solution E g when testing a wind sensor students provided power from a regu lated wall source and mounted the sensor on a gimbal used for testing IMUs in our autonomous vehicles class Students provided a reasonable airflow using a shop fan and looked at the system output on
211. on especially in the box containing the actuators Moreover the fact of sep arating the electronic actuators ensures a perfect seal for the battery and electric circuits 2 2 Use of a Simulator to Design the Rudders To illustrate our strategy for the choice of the different components we present in this chapter the entire strategy for choosing the motors for the rudders of Breizh Spirit 2 The other components and architectural choices have been taken in the same manner for Breizh Spirit 2 and Breizh Spirit 3 One of the most important points for such a small craft is the energy balance because there is a very small area where to lay solar panels The question of energy consumption is more important for the rudder than for the sail You can keep the sail www A TIBOOK ir Breizh Spirit a Reliable Boat for Crossing the Atlantic Ocean 59 degrees Rudder forces Newtons time seconds Fig 3 Evaluation of the force acting on the rudder by manoeuvring Zig Zag simulation fixed for a quite long time while you must control the rudder quasi continuously Thus the question of the design of the entire steering gear is crucial We decided to build a 6 degrees of freedom simulator to evaluate the forces acting on the rudder of Breizh Spirit 2 in order to choose an appropriate actuator The principle of the simulator is thoroughly described in 1 For the current study we do not need such a precise hydrodynamic model Thus we use the very si
212. opti mal for survival as long as the components and electronics could handle the chance of a longer crossing time Current is another factor to consider The velocity is not consistent but climatology shows an average of about 0 5 kt favorable current on both routes This means that light air may not be that detrimental to SOA s progress in a transatlantic crossing because even if the wind is light and the boat is drifting www ATIBOOR ir Route Planning for Micro transat Voyage 191 it will be making progress towards its final destination The voyage will finish off the coast of Antigua Due to the large amount of recreational traffic and anchored vessels in each harbor the plan is to have the SailBot go in to station keeping mode a few miles off the coast until manual control is established Using the same pro cess as the Northern Route the predicted times for the 2500 nm Great Circle Route Southern Route were a fastest time of 27 7 days a slowest time of 45 4 days and a most likely time of 30 5 days The traditional southern route with the dog leg to avoid lighter winds is approximately 200 nm longer and could take anywhere from 27 to over 57 days with the likely time of 32 days only a day and a half longer Both the shorter and longer southern routes are viable 4 2 Final Route Selection Climatological data provided the means for determining likely times for different routes Coastal Pilots provided information for specific departure
213. opyright Law The use of general descriptive names registered names trademarks etc in this publication does not imply even in the absence of a specific statement that such names are exempt from the relevant protective laws and regulations and therefore free for general use Typeset amp Cover Design Scientific Publishing Services Pvt Ltd Chennai India Printed on acid free paper 987654321 springer com www A TIBOOK ir Preface IRSC 2011 is only the fourth in a series of conferences dedicated to robotic sail ing and while still comparatively small we have seen a substantial increase in the number of groups interested in and working on robotic sailboats recently Given that sailing is a fairly old way of locomotion and a high tech sports today it is somewhat surprising that the first competition for autonomous sailboats was proposed as late as 2004 Yet the original objective to autonomously sail across the Atlantic ocean proved to be fairly ambitious and no boat has succeeded so far However this also highlights the complexity of the engineering challenges at hand Sailing depends as much on the physical properties of boat and rig as on the course and route set in the context of changing winds and currents Moreover performance measures do not only include the boat speed but also seaworthiness and robustness of the whole system Hence building a robotic sailboat is a true interdisciplinary project involving naval arc
214. other set of interesting points in the second frame M m m For each interesting point N1 Nn we build a surrounding rectangle of size AxA In the second frame we locate a BxB rectangle at the same place around the interesting point In this stage we start to search the interesting point of the smaller rectangle inside the bigger rectangle If the interesting point is found it will appears in the sequential frame The same search process is implemented for all of the interesting points n n in the same way The target position is determined by the average of all it s interesting points Figure shows the first frame interesting points The red rectangle is marked and all of the interesting points inside it are registered The next step is to build the blue rectangle in the sequential frame and search all of the registered interesting points inside Once 80 percent of the matching is complete we declare those points as relevant and continue searching the next points as a candidate target Fig 7 One Step of Time Correlation Process In the left side the interesting points are marked with red rectangle and in the right side the next time frame introduced with blue rectangle searching the same interesting points from the previous frame www A TIBOOK ir Automatic Obstacle Detection for USV s Navigation Using Vision Sensors 137 Fig 8 Labeling Algorithm Presentation Each label is colored with a different color in this case there
215. ough the deck A Honeywell HMC6343 solid state compass Microchip PIC 18LF4550 microcon troller 2 line LCD screen for debug messages a Newhaven NHD 0220JZ FSPG GBW and Gumstix Connex single board computer were placed in the front com partment and the GPS receiver batteries servos power switch charging cable and USB cable reset switch and re program switch for the PIC were located in the stern compartment To experiment with weight distribution several small plastic bags full of lead shot were taped inside the hull and moved around as required in order to bal ance the hull when it was sat in the water The wing sail featured an attempt to build Deck Hatch For access to power switch charging cable and USB cable USB Cable for reprogramming PIC Sail Servo Futaba S3306 Ballast in keel Power Switch lead shot and Wind P UCC383T 5 5V J regulator for Gumstix and access point EM408 SiRF3 Chipset GPS Magnetic Power linkage to LED rudder Rudder Servo Futaba 2BB MG Charging Bow Light Honeywell HMC6343 Compass D Link DWL G730AP Mini 3x 6 AA 2700mAh Wifi Access NiMH rechargeable Point batteries Gumstix Verdex PIC18LF4550 UCC383T 3 Microcontroller 3 3V regulator for PIC GPS and Fig 2 A digram showing the internal layout of components in MOOPI www A TIBOOK ir MOOP A Miniature Sailing Robot Platform 43
216. partments They use this project as practical work within their studies The team is completed by a student of busi ness administration and engineering He is organizing the needed materials and components The students began their work in February 2011 In May 2011 the mold of the hull of the boat is built We hope we can finish building the boat in July In parallel we are building the controller hardware and the sensors The aim is to have a first version of the whole software in July 2 The Boat Our first sailbot design called FHsailbot is based on an AMYA one meter class specification The size of the hull is limited by the rig especially the size of the available masts We use a profiled aluminum mast with a groove for the sail The maximum size of the mast 2 m leads to a boat length of 1 52 m The second limit ing factor for our first design are the transport possibilities of our cars and trailers To test the control system during the building the boat we equipped an old model boat called Saudade with the controller sensors and actuators So we can test the components and the software independent of the new boat design Figure I shows the hull of the FHsailbot and our test system Saudade Table i compares the param eters of the boats The mold for the hull was shaped from a solid block of closed cell foam using an industrial robot equipped with a milling tool Fig 1 The hull of the FHsailbot left and th
217. ple computers in order to carry out simulations with a wide www A TIBOOK ir 118 C Sauz and M Neal Fig 2 A photograph showing the electronics used in the Hardware in the Loop simulator Numbered labels 1 Telemetry Laptop 2 Tracksail Laptop 3 Wifi Access Point 4 Ar duino 5 Wind Sensor 6 GPS Antenna 7 Compass 8 Gumstix 9 PIC 10 Acuator 11 Current Transducers 12 USB to Serial converters 13 Data Source Switches 14 Parallel Port Connector 15 Cable from Solar Simulation PIC to Arduino 16 Solar Simulation PIC range of experimental parameters It is hoped that this can help reduce the size of the parameter space that needs to be run upon BeagleB and the HIL simulator 3 3 Hardware in the Loop Simulation In order to provide a more realistic simulation an HIL system based upon BeagleB was created This used almost identical hardware to BeagleB placed on a trolley table which can be easily moved around A photograph of this system is shown in Figure 2 An alternative configuration uses a laptop computer running Tracksail to provide mock GPS compass and wind sensor inputs This setup was primarily in tended as a debugging aid unlike a standalone Tracksail simulation this allowed code identical to that which runs on BeagleB to be run on the simulator The abil ity to run long term simulations with an identical code base provides a valuable tool for testing sailing robot control systems for bugs
218. position determines the target angle to be reached by the boat with respect to wind direction We made some assumptions in order to simplify the model First we assumed the rudder is perfectly compensated 1 e the force due to the water current is applied directly on the axis of the rudder As such the wind vane not hindered by the torque of the rudder will align itself with the apparent wind unless the rudder is blocked by reaching its limit angle We also assumed that the apparent wind is equal to the real wind This approximation is valid for slow boats such as ours We also neglected the dynamics of the sail and wind vane since their inertia is small with regards to the magnitude of forces generated by the wind When the rudder is not blocked i e dg g 4x Oomax We have Q c g 3 where c 1 corresponds to the case where the self steering mechanism is on the bow and c 1 when it is on the stern The wind and the regulator are aligned thus t 5 0 are V F 0 4 As such En amp c arg V 7 0 5 When the rudder reaches its limits it becomes blocked In this case we have Og Oginax OF Og Ogna depending on the orientation of the boat relatively to the wind and the sign of In this case one can compute the force F of the wind applied to the wind vane and consequently on the boat We have Q c g 6 www ATIBOOR ir Sailing without Wind Sensor and Other Hardware and Soft
219. presenting a vessel position that is closer than 10m to the desired goal position is considered to be a goal state We use bold to denote vectors and for the scalar product Our heuristic scoring function contoling random node selection is based on the position p of a vessel its velocity vector v and the goal position g __ J O the trajectory of n is not rule compliant ol l h otherwise 3 h p g p g 1 max 0 v p g 4 Any node corresponding to a trajectory that is not rule compliant is assigned a score of zero i e it cannot be selected any more for expansion This ensures that the planner always determines a rule compliant plan For rule compliant nodes the scoring combines distance to the goal with a speed component second term in Equation 4 This term serves to differentiate positions that are similarly close to the goal but in which the vessel is either sailing towards the goal or away from it Random node selection first determines the total score s _ h N of all open nodes N N2 N and then samples a uniformly distributed random number r in the interval 0 s selecting the node N with the smallest value of j such that _h N gt r If anode is selected n random actions are generated and the search f g graph is expanded In order to avoid combinatorial explosion that would occur if continuously ex panding nodes we restrict the size of the set of active nodes which can be further
220. quired criteria 3 i e a sailing boat able to fol low a predefined route in fully energetic and decisional autonomy It was capable of crossing the Bay of Brest in September 2010 as a validation test Nevertheless this boat was largely destroyed during a test between Brest and Morgat To estab lish models of behaviour at sea for crossing the Atlantic Ocean ENSTA Bretagne Richard Leloup Fr d ric Le Pivert S bastien Thomas Gabriel Bouvart Nicolas Douale Henry De Malet Laurent Vienney Yvon Gallou Kostia Roncin Ecole Nationale Sup rieure des Techniques Avanc es ENSTA Bretagne 2 rue Francois Verny 29806 BREST cedex 9 France e mail richard leloup ensta bretagne fr Fr d ric Le Pivert Universit de Bretagne Occidentale UBO 6 Avenue Le Gorgeu BREST France e mail frederic lepivert etudiant univ brest fr www A TIBOOK ir 56 R Leloup et al Fig 1 Breizh Spirit 1 on the Ty Colo Lake decided to build a second boat named Breizh Spirit 2 which is fully instrumented and drawn from studies in CAD Computer Aided Design Finally in view of par ticipating in the Microtransat challenge in November 2010 the students in Naval Architecture ANO and Mechanical Engineering decided to start the design and construction of Breizh Spirit 3 which is 1 4 m long and designed to be able to cross the Atlantic Ocean autonomously based on the experience of the previous two sail boats The main objective for us as w
221. r the MOOPs performance in both of these was not sufficient to be able to run a repeatable experiment When sailing upwind the bow of the boat would point in the correct direction but the actual direction of travel would be sev eral degrees further off the wind resulting in little progress being made Accidental tacks due to small wind shifts or waves were not uncommon and these could easily undo the last 10 minutes of progress During one of the few successful upwind sailing test it took MOOP1 42 minutes to sail 65 m upwind the return journey took only 5 minutes On downwind legs frequent jibes were experienced Attempts were made to perform downwind tacking to control this although to completely prevent unintended jibing the boat needed to sail as much as 45 of the wind www A TIBOOK ir 50 C Sauz and M Neal 3 1 2 Twin Wing Sail Experiments Initial experiments with MOOP3 used a radio control system to steer the boat and test its ability to sail a variety of courses No rudder was used during these experi ments There were two aims to this experiment first to determine if the boat would be stable on all points of sail without a rudder and secondly to test if by setting the sails correctly the boat would settle onto a given point of sail from an arbitrary start ing condition If the latter were true then it would allow for sailing without a wind sensor as the wind direction could be derived simply by setting the sails waiting for
222. r long term routing of autonomous sailboats with an application to the ASV Roboat It is based on the A algorithm and incor porates changing weather conditions by dynamically adapting the underlying rout ing graph We implemented our algorithm in the declarative rule based programing language Constraint Handling Rules CHR 4 A comparison with existing com mercial applications yields considerably shorter computation times for our imple mentation It works with real life wind forecasts takes individual parameters of the sailboat into account and provides a graphical user interface 1 Introduction Autonomous sailboats perform the complex maneuvers of sailing fully automati cally and without human assistance Starting off by calculating the best route based on weather data and going on to independent tacking and jibing autonomous sail boats are able to sail through to any destination Humans merely have to enter the destination coordinates The approach described here is planned to be implemented in the control system of the ASV Roboat an autonomous sailing boat which has been in development by a research team of the Austrian Society for Innovative Computer Sciences INNOC since 2006 So far weather routing on the ASV Roboat relies on locally measured weather data only This is proven to be suitable for short distances respectively short Johannes Langbein Thom Friihwirth Faculty of Engineering and Computer Science Ulm University
223. r sailboats in a declarative language A popular algorithm for path planning in continuous search spaces is the Theta algorithm 1 which also works on a grid of nodes It allows for any angle path planning creating edges to all nodes in sight of the node currently expanded An implementation of this algorithm would provide for a more accurate routing as more angles are considered However this would require many more nodes to be held in memory lead to a higher calculation time and would raise difficulties at choosing the appropriate forecasts for calculating the travel time of an edge Hence we chose the A algorithm with dynamic construction of the routing graph over Theta The D algorithm 13 and its variants are designed for searching in dynami cally changing graphs Their advantage is the fast replanning when costs of edges are modified during the execution of the path which is particularly interesting if a robot continuously collects new information about its environment along the route In robotic sailing however costs are only updated when new forecasts are available which usually happens only every couple of hours Furthermore the D algorithm reconsiders nodes in the closed list which would infer with the dynamic construc tion and deconstruction of the our routing graph cf Section 2 2 For those reasons we opted against the D algorithm in favor of a much simpler implementation and less memory consumption while accepting the
224. rit 3 With regard to the first boat we realised that the weight of the vessel had been underestimated the reserve buoyancy became insufficient That is why the first sailing boat plans 5 and expanded the hull and increased its height while keeping the same form which showed a very good behaviour in manoeuvrability and seakeeping As explained in 5 we chose to inspire the forms of an Open 60 boat from the class IMOCA which is accustomed to facing the Atlantic Ocean at various regatta www ATIBOOR ir 58 R Leloup et al Breizh Spirit 1 had also shown that it is difficult to have a sail boat completely waterproof Various technologies exist to solve this problem For example we can use an automatic bilge pump to drain water 4 However this solution is very energy intensive We can also choose to seal as perfect as possible but this solution is very expensive and not always very reliable 2 7 That is why we chose to make a completely unsinkable sailboat building it from blocks of closed cell foam So even if the boat suffer any collision damaging the hull as we saw with Breizh Spirit 1 there is no risk of sinking because the rest of the hull will keep its own reserve of buoyancy However the integration of electronics and actuators requires waterproof areas That is why we chose to make a boat with three zones with three different levels of sealing e The first level is the area behind the cockpit which is semi sealed protected by
225. rithm 2 1 Modeling Long Term Sailboat Routes For this work we distinguish between long term and short term routing in the fol lowing way Long term routing is the task of finding a sequence of waypoints Xo X longitude and latitude coordinates for a given starting point Xstart and Fig 1 The normalized polar diagram of the ASV Roboat www A TIBOOK ir A Rule Based Approach to Long Term Routing for Autonomous Sailboats 197 time tstart and a given destination point Xgesr Where Xyart X0 Xdest Xn and x is reachable from x _ at sea while taking global weather forecasts into account Short term routing in contrary is the task of finding suitable boat headings to reach the next waypoint given the current local weather conditions 12 Several definitions for the term optimum track in the quote above are possi ble see 12 yet we want to focus on minimizing the arrival time t4est at the point Xdest In order to calculate the arrival time t at any waypoint x we need to take weather data as well as the individual behavior of the sailboat into account 2 1 1 Weather Data As long term routing is typically concerned with distances taking several days or weeks to travel weather forecasts are required to calculate an optimal route Weather forecasts are usually made available in the form of GRIB Gridded Binary files a standardized format to store weather data 14 In a GRIB file wind conditions are represented as
226. rnumber 380245 Caccia M Autonomous Surface Craft prototypes and basic research issues In 14th Mediterranean Conference on Control and Automation MED 2006 pp 1 6 IEEE Los Alamitos 2006 http www engr mun ca bachmayer ENG9095 webpage ASC_USV ASC_survey_paper2006 pdf Cruz N Alves J Ocean sampling and surveillance using autonomous sailboats In IRSC 2008 International Robotic Sailing Conference p 30 2008 http innoc at fileadmin user_upload _temp_ Roboticsailing Proceedings web pdf page 30 www ATIBOOR ir History and Recent Developments in Robotic Sailing 21 21 Elkaim G System identification for precision control of a wingsailed GPS guided cata maran Ph D thesis Standford University 2002 22 Elkaim G The Atlantis project A GPS guided wing sailed autonomous catama ran Navigation 53 4 2006 http citeseerx ist psu edu viewdoc download doi 10 1 1 153 9098 amp rep rep amp type pdf 23 Elkaim G Boyce L C O Experimental Aerodynamic Performance of a SelfTrimming Wing Sail for Autonomous Surface Vehicles In Proc Of the IFAC Conference on Con trol Applications in Marine Systems IFAC CAMS Citeseer 2007 http users soe ucsc edu elkaim Documents PerfCAMS07 pdf f 24 Enab Y Intelligent controller design for the ship steering problem In IEE Proceed ings Control Theory and Applications vol 143 1 pp 17 24 IET 1996 http 1eeexplore ieee org xpl freeabs_all jsp arnum
227. roof switch which isn t reliable 3 2 Hardware and Software Organisation The Algorithm As we said before the boat is using three sensors a compass a GPS and an anemometer With the compass we get the direction of the boat with the GPS the position To be sure there is no error the program make the boat head to the next point after checking four times it is near the buoy as we can see in Fig 7 The anemometer gives the wind direction to the boat In order to be not too sensitive to the variation of wind direction we prefer checking this direction from time to time We do not follow the wind variation and therefore we choose a minimum angle be tween wind and boat direction equal to 50 degrees This angle has been tested and we are sure that the boat may not stop We also do not adjust the rudders exactly for the precise direction we prest five discrete rudder angles chosen with four filter www A TIBOOK ir Breizh Spirit a Reliable Boat for Crossing the Atlantic Ocean 65 Buoy direction Wind direction Angle between sailboat and wind direction Py 4 Angle lt 50 50 lt Angle lt 70 70 lt Angle lt 110 Angle gt 110 Boat Sailboat go Sailboat go Sailboat go direction wind straight on straight on straight on direction 50 g Pull on the Pull on the sail at maximum Sail cross Sail with sail at the wind the wind maximum Fig 8 Direction algorithm angles if the angle between buoy and sailboat
228. rotransat org cited May 22 2011 2 The sailbot competition http www usna edu Users naome phmiller SailBot SailBot ht cited May 22 2011 3 The world robotic sailing championship cited May 22 2011 4 Ammann N Biemann R Hartmann F Hauft C Heinecke I Jauer P Kriiger J Meyer T Bruder R Schlaefer A Towards autonomous one design sailboat racing nav igation communication and collision avoidance In Proceedings of the 3rd International Robotics Sailing Conference International Robotics Sailing Conference June 2010 5 VTS San Francisco http www uscg mil d11 vtssf cited May 22 2011 4 Test system Windows XP Intel Pentium M 1 86 GHz 2 GByte RAM www A TIBOOK ir 166 N Ammann et al 6 National Oceanic and Atmospheric Administration National Weather Service National Data Buoy Center USCG NOAA Automatic Identification System AIS on Data Buoys and C MAN Stations 2004 7 The RoboCup Federation User Manual RoboCup Soccer Server for Soccer Server Version 7 07 and later 2001 8 Sauze C Neal M A raycast approach to collision avoidance in sailing robots In 3rd International Robotic Sailing Conference Kingston Ontario Canada pp 26 33 June 2010 9 Stelzer R Jafarmadar K Hassler H Charwot R A reactive approach to colli sion avoidance in autonomous sailing In 3rd International Robotic Sailing Conference Kingston Ontario Canada pp 34 40 June 2010 ww
229. rough a simplistic mechanism which specifies upper and lower bounds for the wind direction and the extent to which it will vary We have observed a number of key differences in behaviour between Tracksail and real sailing robots In particular the response rates to rudder and sail movements of the simulated boat are much faster than real boats and there is no lag between initiating an actuator movement and it completing Despite these limitations in pre vious work using Tracksail to develop power management strategies similar trends were exhibited between Tracksail and real sailing robots 10 Another possible approach for simulation might be to adapt commercial games such as Virtual Sailor 2 or Sail Simulator 1 These both offer simulations of a variety of boats waves varying wind conditions and appear to have far more realistic physics models than Tracksail However being closed source games they are not intended for sailing robot simulation and may not easily be adaptable for this purpose One possible means to achieve this might be to communicate with them through the network protocols intended to provide multi player support There is relatively little other scientific literature regarding the simulation of sail ing robots Briere 2008 4 produced a MatLAB simulation of a 2 4 m long sailing robot This simulation included both consideration of the aerodynamic and hydro dynamic aspects of the boat It also presented an software interfac
230. rrents The twin wing sail designs have shown themselves to be more stable on downwind courses and offer the potential of rud derless sailing or at the very least that the rudder and sail can be used cooperatively Despite the name Miniature Ocean Observation Platform it remains unlikely that a MOOP is likely to be able to complete any serious ocean sailing unless it is sailing downwind and with the currents This might for example be achievable between the Canary Islands and the Caribbean but it is unlikely that a MOOP will be able to sail from the UK or Ireland to the Canary Islands which rules out an entry into the Microtransat Challenge following its 2010 2011 rules When compared with other small boats such as the MicroMagic boats used by the University of Liibeck or Breizh Spirit built by ENSTA Bretange upwind performance of the MOOPs is much poorer This is believed to be due to the drag caused by the wide keel design If serious ocean sailing against tides and currents is to be achieved then a redesign of the hull shape may be required Further sea trials of the MOOPs are required Reductions in power consumption are also required for prolonged sailing One possibility is to use a self correcting wing sail similar to that used by Elkaim and Boyce 8 where a small rudder on the wing sail is used to control its angle once the rudder is set the sail will maintain its position with respect to the wind without the need for an active control
231. rystwyth UK featuring a hysteresis condition to avoid permanent switching between two adjacent positions 17 Rea sons for a reduced number of sail positions are to save power on the sail actuator and to extend the lifetime of the sail gear A state machine to allow for special sail trim during manoeuvres such as tack and jibe has been implemented on D umling Uni versity of Liibeck Germany Avalon Swiss Federal Institute of Technology Zurich Switzerland and Boat ISAE France 25 14 17 A method which does not directly calculate a sail position based on wind data was published by Stelzer et al 49 It firstly determines a desired heel for the boat out of speed and direction of the apparent wind A feedback loop implemented as a Mamdani type fuzzy inference system then controls the sail position towards this heel value All methods described above can basically be applied to both conventional and wing sails For the latter a further control method has been published namely a self trimming wing sail A self trimming arrangement typically consists of a wing sail vertically mounted in bearings that allow free rotation A smaller wing called tail is usually mounted just behind the main wing see figure An aircraft uses tails to control the exact angle of attack of its wings Similarly the tail on a wing sail system is able to control the thrust obtained from the wind and will automat ically take into account any changes in wind direction
232. s The server offers various interfaces for the communication with the clients us ing different serialization methods Java object streams XML and a custom string based protocol All of these serialization methods have in common that they use TCP sockets as transport protocol 4 2 Clients Clients do not necessarily need to represent a physical boat it is possible to simulate artificial traffic by adding virtual boats controlled by programs Another example is our graphical user interface which connects to the world server as an ordinary www A TIBOOK ir Collision Avoidance Using a World Server 163 client too Another client could work as an artificial race judge checking for rule violations The world server is implemented in pure Java running on virtually any operating system Likewise the Java based client is platform independent To demonstrate the flexibility of the World Server we have also implemented and tested a small client written in C using the World Server s custom string based protocol This protocol is based on ASCII coded strings and directly writing them to or reading them from socket streams The first element of each message is an identifier ranging from 1 to 5 followed by the line separator n or n r respectively The meanings of these identifiers are 1 sent by the World Server to a client followed by the objects registered at the server followed by a final line separator indicating the e
233. s www A TIBOOK ir Breizh Spirit a Reliable Boat for Crossing the Atlantic Ocean 69 been entirely rebuilt to ensure its durability Now the Microtransat project arouses a lot of interest from industry and research Therefore we will continue to develop various research programs on Breizh Spirit 2 The eventual aim being to cross the Atlantic some tests will still be made with Breizh Spirit 3 drawing conclusions from the misadventures of Breizh Spirit 1 Indeed despite of all the work as can be seen from the Table 3 it remains to test the endurance of the boat but also its behaviour in wind less than 10 knots and more than 30 knots upwind With these various experiences we hope to be able to compete in September 2011 and meet the Microtransat challenge Acknowledgements Thanks to DCNS a company specialized in naval shipbuilding to sup port the project Breizh Spirit Thanks to all who contributed to the realisation of the Breizh Spirit Project Especially Michel Jaffres and Bruno Aizier References 1 Roncin K Kobus J M Dynamic simulation of two sailing boats in match racing Sports Engineering 7 3 139 152 2004 2 Briere Y IBOAT An autonomous robot for long term offshore operation In Proceed ings of MELECON 2008 The 14th IEEE Mediterranean Electrotechnical Conference May 5 7 2008 3 Briere Y Bastianelli F Gagneul M Cormerais P Microtransat challenge In CETSIS 2005 Nancy France vol 5 2 200
234. s only possible with the help of many people We are grateful to all of them particularly to Petra Ro kopf for her assistance with the conference location and to our student co organizers who have worked tirelessly to get everything arranged We also thank our sponsors and partners without whom this conference would have been infeasible Alexander Schlaefer is very appreciative for the patience and tolerance of Achim Schweikard who supported the idea to organize IRSC WRSC in L beck Ole Blaurock would like to express his appreciation to the robotic sailing team at the Fachhochschule L beck which started only in the beginning of the year and yet managed to activate numerous colleagues enabling the participation in WRSC 2011 L beck Alexander Schlaefer August 2011 Ole Blaurock www ATIBOOR ir Organization General Chair Alexander Schlaefer Co Chair Uwe Krohn Proceedings Alexander Schlaefer Ole Blaurock Secretariat Cornelia Rieckhoff Student Co Organizers Niko Ammann Lenka Hanesov Florian Hartmann Philipp Jauer Julia Kr ger Tobias Meyer Program Committee Jos C Alves Ralf Bruder Nuno A Cruz Erik Maehle Benedita Malheiro Paul Miller Mark J Neal Cedric Pradalier Colin Sauze Roland Stelzer University of Liibeck Germany University of Applied Sciences L beck Germany University of Liibeck Germany University of Applied Sciences Liibeck Germany University of Liibeck Germany Un
235. s potential routes for initial planning to determine expected times and narrow the potential field of routes and an off the shelf sailing yacht route optimization program for routing immediately prior to or during the crossing The long term method used climatological data from Pilot Charts A Velocity Prediction Program VPP was used to predict the boat s perfor mance Calcualted passage times included the shortest longest and most likely A northern and two southern routes with similar risk were identified 1 Introduction Preparing for an ocean crossing involves extensive planning and preparation regard less of the vessel type Making a crossing of the Atlantic Ocean fully autonomously is the goal of the MicroTransat Competiton and to be successful not only is a good boat a requirement the route selected must have a high probability of success and the boat must be designed to meet the expected conditions for the selected route This paper discusses the research that went into the route planning for the Naval Academy s 2011 SailBot Spirit of Annapolis which was designed and built by undergraduate students The long range planning used climatological data and the oretical boat velocities to estimate possible crossing times for different routes Var ious factors influencing the transits were studied to provide routes with the highest Peter Gibbons Neff Paul Miller United States Naval Academy Annapolis Maryland USA e mail Of c112
236. sails are shifted individually to keep the boat stable k l typically operating both sails with a sheet which apart from a small offset for the jib cannot be adjusted independently We initially used to move both sails proportional to the apparent wind angle e g during a tack both sails moved synchronously with the apparent wind However we quickly noticed that this made tacking in stronger winds and against short and steep waves of approximately 0 1m almost impossi ble as the boat would quickly come to a complete stop and loose rudder control Slacking the jib before initiating the tack and maintaining an offset until the boat is on a close hauled course on the new tack improved the tacking substantially com pare Figure 6h f Similarly it is virtually impossible to have the boat fall off in stronger wind unless the main is slacked first as shown in Figure 6k n This high lights that individual sail control and the ability to back the sails can help control the boat 5 Results Our current design is intended as a prototype for the class and we expect future improvements to virtually all aspects Here we summarize some basic results e g regarding the robustness of the sensors and the control board and with respect to the sailing performance www A TIBOOK ir A New Class for Robotic Sailing The Robotic Racing Micro Magic 79 5 1 Robustness As size was one criteria when selecting the Micro Magic as the basis for an one desi
237. situ reconfiguration Note that the full set of functions is not defined here but there are several func tions described above that would not be clear from the basic SailBot task Students must fully enumerate the set of functions while avoiding solution bias once the full set of objectives is defined The set of functions things that the system must do is given by the implicit functions above as well as the following 1 Allow configuration of task the software or hardware must allow a user to select from a suite of tasks station keeping match racing etc and configure the parameters of the task maximum allowable deviation GPS coordinates etc 2 Compute sailing commands the system must be able to determine appropri ate sailing commands to control the available surfaces typically for our ves sels the rudder and the main sail 3 Carry out sailing commands the system must execute sailing commands meaning that it must actuate the surfaces at the appropriate time and to the ap propriate position 4 Measure appropriate vessel states the system must measure data as required by the sailing command generation subsystem The exact nature of the data re quired should not be specified at this point 5 Avoid obstacles the system must recognize and avoid navigation obstacles including other vessels and structures that might damage the vessel 6 Provide remote override the system must allow a human to remotely take con
238. st identical while the route computed by Sailplanner is somewhat different There are two reasons this difference might stem from Firstly Sailplanner uses a different polar diagram which could not be changed in the demo version available to us Secondly Sailplan Table 1 Comparison of wall clock computation time required for routing in seconds SailFast Sailplanner Sailplanner Roboat router Roboat router 6 hours med high ultra high 20km 30 km Route 1 109 40 254 10 6 Route 2 79 a 224 9 5 www ATIBOOR ir 202 J Langbein R Stelzer and T Fr hwirth ner was run with wind data from WeatherTech as it does not allow the import of GRIB files Our algorithm and SailFast were both run with the same GRIB file from lsaildocs comisince the data from WeatherTech is not freely available In academic research several methods have been proposed for sailboat rout ing see for an overview A stochastic method for long term routing based on dynamic programming was presented by Philpott and Allsopp 7 while meth ods from operations research were used by Papadakis and Perakis 6 Recent work bythe AVALON team 3 uses a routing algorithm similar to ours however they do not include weather forecasts in their calculations To our knowledge none of the aforementioned approaches have been implemented in a rule based language like our algorithm and there has not been a published implementation of a long term weather routing algorithm fo
239. sts of an unstayed mast carrying a main and jib see Fig 2 The main boom extends forward of the mast the mast passes through the boom to the tack of the jib The main and jib are sized so that the force from the mainsail is slightly higher than that from the jib That is the combined center of effort is just behind the mast Therefore the load on the sheets is reduced by more than 50 compared to a conventional rig due to the balanced distribution of the sail load caused by wind 25 Balanced rigs have been used on the autonomous sailing boats Avalon and Boat 14 Furthermore most of the rigid wing sails mentioned above can be con sidered to be balanced rigs 5 Rigging is the mechanical sailing apparatus attached to the hull in order to move the boat as a whole This includes cordage sails and spars masts and other solid objects sails are attached to www A TIBOOK ir 8 R Stelzer and K Jafarmadar Fig 2 Balanced rig example the combined center of effort is just behind the mast 2 2 2 Sail Control Strategies Most sail control strategies published for autonomous sailing boats rely on locally measured apparent wind data only 1 17 25 While many of them have a virtually infinite number of sail positions limited just by the resolution of the actuator or the used data types and therefore allow smooth sail control just 10 discrete sail position are used on MOOP University of Abe
240. t 9508 the rules it will be possible to experiment with different rigs and methods to actuate the sails The eighth rule has been added to promote collision avoidance which we identified as another key problem particularly in fleet races with similar boats According to the class rules the boats are not required to be completely autonomous 1 e the algorithms controling the boat can be executed remotely Ad vantages include less weight and power consumption onboard the use of high per formance standard computer hardware and high level programming languages and simplified debugging during test runs 3 Hardware Assembling the Graupner kit is rather simple and we refer to the Graupner manual 6 and only briefly summarize our experience building the boat Particularly we will present a few modifcations to drive the sails and then focus on electronics and sensors More information is available online 10 3 1 Boat The mechanical dimensions of the Micro Magic are shown in Table 1 Hull keel and rudder are mounted according to the Graupner instructions However the original Graupner design uses sheets to control the sails and while the jib is attached to a boom the latter is connected to the deck by a small string This complicates adding sensors to measure the sail angles and we therefore opted for a direct link between sail servos and booms The jib boom is mounted on a small vertical pole close to the bow and both servos are mounted
241. t of metrics for the primary objectives of the SailBot system with the point values for each level of performance 1 Fast given a change in environment state or task configuration the system with compute and reach a new course in 4 points less than 5 seconds 3 points between 5 and 7 seconds 2 points between 7 and 9 seconds 1 point between 9 and 10 seconds 0 points greater than 10 seconds 2 Accurate passes within a specified distance on a specified heading of a tar get GPS point corresponds to scoring buoys in races The best score is achieved when passing inside the start marks which are 3m apart There are further marks an additional 3m beyond those start points 4 points passes within Im of target 3 points passes within 1 1 5m of target 2 points passes within 1 5 3m of target 1 point passes within 3 4 5m of target 0 points greater than 4 5m from target 3 Energy efficient the system must provide energy for endurance trials The scoring rubric is battery life in hours divided by 2 with a max score of 4 8 hours of battery is desirable 4 Robust this metric scores based on 1 point binary decisions for a total score of O 4 1 system automatically reboots and restarts configured task when power is interrupted 1 system components are watertight to 6 1 system components are secured against shock mean time between failures in normal operation of greater than one day 1 system can function in a
242. ta or path planning run asyn chronously in separate threads Another advantage of the onshore implementation is simple debugging and prototyping e g the software can be changed and compiled without even restarting the boat Moreover essentially any high level language can be used for development and we are currently having prototypes implemented in Java and C Although we are just starting to analyze log files from test runs we have iden tified an interesting aspect of sail control The conventional racing Micro Magic is Fig 5 The sensor data processing loop operating at 50 Hz green with the command queue executed at 5Hz Frequency and processing time for sending and receiving messages are shown inpurple and yellow respectively The yellow signal also indicates that the external frequency could be increased to approximately 20 Hz www A TIBOOK ir 78 A Schlaefer et al Fig 6 The figure illustrates how jib and main sail are moved independently during maneu vers On the left side a typical tack from a close hauled on starboard tack to a close hauled on port tack is shown a f Note that the jib is opened before turning the rudder b which is crucial to power through short waves The jib is slack in c and starts backing in d to finally push the boat onto its new course e f On the right side a jibe from beam reach to beam reach is illustrated g n The main sail is opened before initiating the jibe h and the
243. tability of automatic steered bodies Journal of American Society of Naval Engineers 34 2 280 309 1922 33 Multirig Multirig website 2009 http www multirig com the_ balestron_rig htm accessed April 16 2009 34 Neal M A hardware proof of concept of a sailing robot for ocean observation IEEE Journal of Oceanic Engineering 31 2 462 469 2006 http citeseerx ist psu edu viewdoc download doi 10 1 1 103 1486 amp rep repl amp type pdf 35 Neal M Sauze C Thomas B Alves J Technologies for Autonomous Sailing Wings and Wind Sensors In Proceedings of the 2nd IRSC Matosinhos Portugal July 6 12 pp 23 30 2009 http cadair aber ac uk dspace bitstream 2160 3103 2 1IRSC pdt 36 Polkinghorne M Roberts G Burns R Winwood D The implementation of fixed rulebase fuzzy logic to the control of small surface ships Control Engineering Prac tice 3 3 321 328 1995 http 202 114 89 60 resource pdf 2231 paf 37 Roberts G Trends in marine control systems Annual reviews in control 32 2 263 269 2008 http 202 114 89 60 resource pdf 2236 pdf www ATIBOOR ir 22 38 39 40 41 42 43 44 45 46 47 48 49 50 51 32 R Stelzer and K Jafarmadar RoboCup Official robocup website Objective 2011 http www robocup org about robocup objective accessed May 1 2011 Rynne P von Ellenrieder K Unmanned autonomous sailing Current status a
244. tations can be incor porated in logic based agent control 3 Such techniques tackle planning only on the level of abstract actions though Navigation rules in particular codes of prac tice for sailing can also be captured in Fuzzy representations 16 but they lack the formal semantics that allow abstract processes to reason about the consistency of actions possible with qualitative representations 6 Navigating by qualitative rules requires bridging from abstract spatial relations to concrete control parameters needed by the actuators of the robotic system Thus symbolic reasoning needs to be linked to control parameter selection The example of tacking a complex turn maneuver in sailing see Fig 1 illustrates the difficulty of this integration Tack ing requires several preconditions to be met in order to perform the maneuver for example enough free maneuver space needs to be available The amount of space required depends on the specific physical context like wind initial vessel speed in ertia of the vessel etc If the initial speed of the vessel is too slow tacking fails It appears to be difficult to come up with an abstract definition of tacking that is pre cise enough to represent exactly those situations in which the maneuver is possible For example overestimating the space requirements may inhibit planning to iden tify situations in which the action can be performed underestimating it may lead to accidents Thus the appli
245. ters and to adjust sail and rudder position Extensive research has been undertaken on semi autonomous systems where just a subset of the functionality of a robotic sailing boat is covered The history of self steering gears and automatic sail control will be discussed independently in the following sections Afterwards a separate section shows history and recent research projects in completely autonomous sailing 2 History of Robotic Sailing 2 1 Self steering Gear Historically the first task to be automated was the governing of the rudder A self steering gear 1s an equipment used on ships and boats to maintain a chosen course without constant human action Self steering gear is also referred to as autopilot or autoheln Basically the different forms of self steering gears can be divided into two categories mechanical and electronic 2 1 1 Mechanical Self Steering Fishermen who bind the rudder or tiller of their boat in a fixed position to produce an optimal course can be seen as a first approach to a mechanical self steering system 371 A more sophisticated mechanical approach is the wind vane developed first by Herbert Hasler 1914 1987 who is known as one of the fathers of single handed sailing Wind vanes are now sold by a number of manufacturers but most share Autohelm is a Raymarine trademark but often used generically Single handed sailing is sailing with only one crew member The term is usually used with reference
246. the aim to build a self learning autopilot for a single handed sailing yacht Agent technology machine learning data mining and rule based reasoning have been combined to a system which be came commercially available after five years of development 2 A rudderless approach on automatic heading control of a sailing boat was pre sented by Benatar et al 8 They have shown that control of a rudderless boat with two sails can be achieved by coordinating the two sails for propulsion and turning 2 2 Automatic Sail Control While extensive research has been carried out on automatic steering devices au tomatic sail trim is a more recent idea and not yet well covered by scientific publications www A TIBOOK ir History and Recent Developments in Robotic Sailing 7 2 2 1 Rigging and Sails So far several different rigeined have been used on robotic sailing boats They can be characterised according the following criteria e Material and shape traditional fabric sail or rigid wing sail e Rigging Balanced or unbalanced In history of sailing which lasts several thousands of years a large variety of dif ferent sail shapes and technologies have been used Virtually all boats apart from the recent sailing history used conventional fabric sails This form of sails have some advantageous properties especially when controlled by a human sailor This includes the easy way of reefing repairing or that shape and camber can be altered by s
247. the one used on big boats since the applied forces are smaller To overcome this problem it is possible to use bearings but because of salt corrosion one have to use special bearings such as plastic bearings with glass balls We propose the following design of the bow self steering system using the fact that a round buoy rotates smoothly on sea water as seen in Figure 6 Figure represents our first attempt to construct the self steering device The buoy is constructed using extruded polystyrene covered with fiberglass and resin for www A TIBOOK ir Sailing without Wind Sensor and Other Hardware and Software Innovations 33 wind vane I front of the boat Pa 2 eed rudder Fig 5 The kinematic diagram of the bow wind vane self steering device wind vane servo winch Fig 7 The actual design of the first prototype using CATIA computer assisted prototyping software www A TIBOOK ir 34 J Sliwka et al Fig 8 This figure illustrates a sailboat navigating from a waypoint A to a waypoint B more robustness A PVC sheet is sandwiched between two parts of the polystyrene The PVC sheet supports a servowinch and a box for the electronics 3 Navigating in Straight Line We consider a robot boat navigating from a waypoint A to waypoint B as seen in Figure 8 If the waypoint B is not upwind the strategy is to try to regulate the boat such as it always targets waypoint B Denote by M x y the positi on of the robot
248. the wind vane self steering device is that this navigation is relative to the direction of the wind As a consequence the wind sensor is no longer necessary In this paper we also presented few algorithms to navigate between waypoints The real tests were performed using our recently build test platform presented in the last section References 1 http www uovehicles com 2 Bertram V Unmanned surface vehicles a survey Skibsteknisk Selskab Copenhagen Denmark 2008 3 Briere Y Bastianelli F Gagneul M Cormerais P Challenge microtransat In CETSIS 2005 Nancy France 2007 4 Forthmann P C Self steering under sail Windpilot 5 Jaulin L Modelisation et commande d un bateau a voile In CIFA2004 Conference Internationale Francophone d Automatique CDROM Douz Tunisie 2004 6 Schlaefer A Bruder R Stender B Model sailboats as a testbed for artificial intelli gence methods In IRSC 2009 pp 37 42 2009 7 Sauze C Neal M An autonomous sailing robot for ocean observation In TAROS pp 190 197 2006 www A TIBOOK ir 38 10 11 J Sliwka et al Sauze C Neal M Design considerations for sailing robots performing long term au tonomous oceanography In IRSC 2008 pp 21 29 2008 Sauze C Neal M Ocean sampling and surveillance using autonomous sailboats In IRSC 2008 pp 30 36 2008 Sliwka J Reilhac P H Leloup R Crepier P Malet H D Sittaramane P
249. tical miles at an average speed of 3 knots and it reached several times a maximum of 5 6 knots WPT4 Swensea Vale Fig 10 Track of the Brest Morgat test On the center of the figure we can see that Breizh Spirit 1 couldn t tack upwind on 30 knots wind speed www A TIBOOK ir 68 R Leloup et al The following year was dedicated to enhance reliability and to develop a system validation and verification procedures for each sensor and actuator It remained to test the behaviour of the vessel at strong sea state The vicinity of the Ushant Island Ouessant in French often offers extremely hard sea conditions especially when the wind blows against the current Thus it is an excellent playground for testing seakeeping ability and endurance of our platform Two attempts were conducted and permitted to detect some weakness in our electronic and mechanical systems For the second attempt around Us Island Breizh Spirit 1 travelled 12 nautical miles upwind tacking in waves up to 2 to 3m Its behaviour was very satisfactory beyond our own hopes We considered then that only the endurance function re mained to validate for crossing the Atlantic Ocean For the third attempt the meteorological forecast was so uncertain that we chose only to go round the peninsula of Crozon from Brest to Morgat Fig 10 shows the track recorded before the crash of Breizh Spirit 1 One can see on course that there is a Strong drift of the boat between the way poi
250. tions are to be distinguished In our work we use m 4 Fig 4 For each of the two related oriented points m lines are used to parti tion the plane into 2m planar and 2m linear regions Direction O is aligned with the orientation of the point If two positions coincide so called same relations occur In these cases a number s denotes the direction in which B is oriented as seen from A Fig 4 b written as A ZsB OPRA provides no atomic front or back region as it is needed to represent head on for example Such region can easily we described as disjunctions of finer grained OPR Asg relations To ease notation we define OPR Aj relations as a shorthand notation for such disjunctions Geometrically OPR AZ relations can be obtained as follows we rotate the segment boarders of OPR A by half of the an gular resolution 1 e by 5 360 22 5 and join the linear regions with the planar ones In Fig 6 we depict an example of relation a www ATIBOOR ir 146 D Wolter F Dylla and A Kreutzmann aV a head on A 4x9 B b crossing A e B c port side wind A 425 Awind Fig 5 Two sailing vessels in a head on and a crossing situation and Fig 6 An example re the advised collision avoidance behavior lation for OPRAJ A 4x9 B 4 Formalizing Navigation Rules The basis for our rule formalization is formed by parts of the International Regu lations for Avoiding Collisions at Sea COLREG
251. tions thereof have received considerable attention for rudder control on ships Var ious publications have shown the suitability of FL for rudder control 55 1 56 149 Polkinghorne et al furthermore made a comparison to its conventional PID controlled equivalent The experiments have shown a much smoother rudder action for the FL controller Stelzer et al again demonstrated a reasonable perfor mance of a FL controlled rudder even during tacking and j bing Adaptive FL controllers have been presented as a promising approach to com bine expert s knowledge and new experiences automatically In Velagic et al a Sugeno type fuzzy inference system is combined with a feedback loop to adjust scaling factors of the base fuzzy system For the mentioned FL approaches human expert s knowledge must be known a priori to design the fuzzy rule set In contrast Layne and Passino 26 published a learning control algorithm which automatically generates the fuzzy controller s knowledge base on line as new information on how to control the ship is gathered Other examples of adaptive rudder control systems are based on artificial neural networks Both Enab and Burns aim to pro vide an ANN based system that can adapt its parameters towards optimal perfor mance over a range of conditions without the need of manual adjustments An ambitious machine learning approach to automatic rudder control was the Ro boSail project 51 The project started in 1997 with
252. to be sealed Moreover it is difficult to use because of its high consumption Therefore we chose the servo motor HSB 805 BB from Hitec which seems to be the best for Breizh Spirit 2 2 4 Construction of the Hull from a Digital Model For Breizh Spirit 2 and Breizh Spirit 3 we decided to make a digital model of the boat and its components to facilitate the design of elements of the hull In this section we will focus primarily on the achieving of Breizh Spirit 3 www A TIBOOK ir Breizh Spirit a Reliable Boat for Crossing the Atlantic Ocean 61 Longitudinal Bulkheads Transversal Bulkheads Semi Waterproof Zone Fig 4 Digital model of the hull As we have seen before using foam blocks has an obvious advantage in terms of unsinkability The second advantage is that we do not have to make a mold as was done for Breizh Spirit 1 Using the digital model allowed us to obtain a SperfectT hull shape However the choice to build this boat from foam blocks machined with a milling machine required us to fully draw each block We used the software devel oped by Dassault System Catia which allows us to fully define the geometry of the hull form from a surface recovered from the naval architecture software DelftShip Once the hull is designed Catia generates all the programs that are then given to the milling machine for the machining of the foam The definition of the blocks has been conditioned by different parameters e The movemen
253. to char acterize the targets and to distinguish between the target s and the environment s characteristic avoiding false alarms 3 4 Heuristic methods were introduced in the early 1980s based on threshold gradient in the image The threshold was determined by the contrast of an object with its local background 3 The segmentation part of such algorithms is based on standard edge operator using closing shape algorithms and filling steps 5 6 7 8 Most of the previous works were studied in aerial and ground environments with out considering special phenomenas in marine environments such as waves clutter vehicle s stability etc For the first time our algorithm deals with ATD algorithms for USV s motion planning and automatic decision models using COTS sensors The algorithm based on common video format as input and computes targets loca tions around the vehicle 2 General Description Our new algorithm gets an input video in one of the common formats avi mpeg mov etc starts a few processing steps and eventually finds all the targets in the specific image The main steps as shown in Figure lJare Read input image from the COTS device Resize and convert the image into gray scale reducing CPU time 3 Reduce search space for targets by preforming initial cleaning and horizon identification 4 Learn the sea pattern using a co occurrence matrix 5 Morphologic cleaning which cleans the image after distinguis
254. to ocean and long distance sailing www A TIBOOK ir History and Recent Developments in Robotic Sailing 5 Fig 1 Example for a wind vane with trim tab on main rudder from 41 the same principle The device consists of a wind vane secured at the stern of the yacht which is connected to the rudder respectively a trim tab on the rudder via a system of ropes pulleys and servos see figure l When the angle of the apparent wind changes this change is registered by the air vane which activates the steering device to return the boat to the selected point of saiff Wind vane self steering does not steer a constant compass course but a constant point of sail 2 1 2 Electronic Self steering Electronic self steering controls the rudder movement by electronics based on var ious sensor input values At least a compass is necessary additional sensors can deliver wind direction or GPS position in order to calculate a heading towards a given target waypoint Substantial progress toward automatic steering was based on the invention of electronic gyrocompasses The earliest known gyroscope like instrument was made by German Johann Bohnenberger who first wrote about it in 1817 12 Accord ing to Bennett 9 and Roberts the major contributions to the development of a practical automatic steering system were made by Sperry Gyroscope Com pany Elmer Sperry developed his first automatic ship steering mechanism in 1911 46 Sperry s gyropilot was
255. trol of all sailing functions in order to maintain safety 7 Transmit data the system must provide telemetry to the shore indicating all data gathered and computed sailing commands 8 Provide power the system must provide power sufficient for all subsystems for the specified operational period 1 2 2 Demonstration Plan and Metrics The set of design objectives and functions above are the hallmarks of a good de sign but they are vague and open to interpretation Students must be provided with a method by which they can test and validate their design in a rigorous and meaningful way It is here than many competition based projects run into trouble Specifically it is inappropriate to design to a metric that cannot be evaluated www A TIBOOK ir A Systems Engineering Approach to the Development of an Autonomous Sailing Vessel 91 during the design process If victory at the competition is the only performance goal it will be impossible for the students to carry out the iterative design of the individual components and subsystems that is required by a quality design proc ess As such students pursuing SailBot at USNA must define a set of metrics by which their system can be assessed before the competition This set of metrics must be driven by the tasks of the competition and realistic performance levels must be enumerated we use a 0 4 system with 4 being the best result and O being unacceptable The following describe a full se
256. trol system is implemented on an MPC21P9 industrial computer running Linux 3 3 6 United States Coast Guard Academy USCGA Intuition was developed by the United States Coast Guard Academy and participated in SailBot WRSC 2010 The USCGA s 2 m monohull design features a conventional 16 Jan Sliwka 17 INNOC 18 INNOC 19 Patrick Moser 20 http www kontron com www ATIBOOR ir History and Recent Developments in Robotic Sailing 17 sloop rig with a sail area of 1 7 m The control system is implemented on an ISIS PC104 single board computer running Windows XPe and MATLAB 17 3 3 7 United States Naval Academy USNA USNA began their activities in 2007 They participated with their boat First Time Figure 76 in SailBot 2008 In 2009 they entered Luce Canon Figure KOA in SailBot and WRSC In 2010 they participated with Gill the Boat Figure HOA in SailBot WRSC The USNA team comprising undergraduate Naval Architects and Systems Engineers designs and builds new sailing vessels each year by continuously improving upon previous year s boats Their designs use a custom built single hull with a length of 2 m and a conventional sloop rig with a sail area of about 3 m 3 3 8 University of Aberystwyth The team around Mark Neal and Colin Sauz built multiple sailing robots varying in length between 50 cm and 3 5 m with the aim of performing long term autonomous missions for oceanographic monitoring AROO Figure EFD was
257. ts and courses of the milling machine e The position of the partitions e The integration of components e The thickness of the foam raw Indeed because the movement of the machine is limited to 0 5m in length and width and 0 3m in height we were forced to carry out at least 6 blocks being given the dimensions of our hull Otherwise one of the major points during the design of a boat is the consideration of the bending moment and the torsion moment due to the keel and shrouds That s why we built the hull with longitudinal divisions in green www A TIBOOK ir 62 R Leloup et al Closed Cells Foam Reinforced with Fiberglass epoxy Marine Plywood i Watertight electronics box Fig 5 Breizh Spirit 3 s hull on the Fig 4 to withstand the bending moment due to wave action and tension in the rig The transverse bulkheads in red on the Fig 4 withstand the torque moment Finally we got a shell constructed from 13 blocks machined with the milling machine The cutting of a block of the hull was made from a rectangle of foam Once the blocks were cut they were assembled to obtain a stiffened shell with walls made from fibreglass impregnated with epoxy The boat was finally covered with an external skin using a vacuum pump to impregn fiber with epoxy to creat a composite as we can see in Fig 5 3 Development of Reliable and Robust Electronics 3 1 Reliable Electronics To optimize power consumption on board the electro
258. ture research Our algorithm is limited to sensor noises and false alarm dealing with very cluttered environments The algorithm has been tested with stabilize sensor on the platform which simplify horizon line recognition In case of not stabilized sensor or panoramic image integrating several video streams we expect a very limited success rate Another well know problem in marine environments is related to sun effects The algorithm might classify sun light effects as targets which will cause false alarms This challenge should be treated in our further research for more accurate ATD abilities for accurate USV trajectory planning The highlight of the algorithm is the simple concept that can be use on COTS sensor without special hardware and as far as we know this specific issue was not yet studied extensively in marine environments The algorithm is based on simple basic algorithms from the image processing world which are suitable for real time application We presented a new algorithm to acquire identify and track obstacles from USV systems using COTS sensor for autonomic navigation The algorithm is based on previous image processing filters and algorithms and been been adapted to the ma rine environment challenges The algorithm was successfully tested on real time video from USV systems and can be apply in real time applications dealing with CPU time constraints Further research directions include light effects and not stabilized syst
259. turn rate approximates that of the radar Furthermore on our class of boats a 4 meter sailing vessel the radar will be mounted at quite a low height of about 1 m and will need a heeling compensation either mechanically or algorithmically e g removing reconstructing bad parts of scan This may require further test with different radar antenna angels relative to reference plane ground sea Therefore we expect a fairly limited range in compar ison with the technically possible range References 1 Almeida C Franco T Ferreira H Martins A Santos R Almeida J M Car valho J Silva E Radar based collision detection developments on USV ROAZ II In OCEANS 2009 EUROPE pp 1 6 IEEE Los Alamitos 2009 http 1eeexplore ieee org xpl freeabs_all jsp arnumber 5278238 ISBN 978 1 4244 2522 8 www A TIBOOK ir A Digital Interface for a Lowrance Broadband Radar 181 2 Austrian Association for Innovative Computer Science INNOC ASV Roboat http www roboat at 3 Benjamin M R Leonard J J Curcio J A Newman P M A method for protocol based collision avoidance between autonomous marine surface craft Journal of Field Robotics 2006 4 Cheshire S Aboba B Guttman E RFC 3927 Dynamic Configuration of IPv4 Link Local Addresses 2005 5 Holbrook H Cain B Haberman B RFC 4604 Using Internet Group Manage ment Protocol Version 3 IGMPv3 and Multicast Listener Discovery Protocol Version 2 ML
260. uggle to turn more than 127 degrees this caused problems rounding waypoint 1 and waypoint 2 During the upwind leg between waypoints 2 and 3 BeagleB tacked several times this is despite the tacking algorithm being used which attempted to make only one tack This was due to noise from the wind sensor causing the control system to believe it was now able to sail directly to the waypoint but after tacking the wind direction would shift The lower levels of noise in the HIL simulation and the Tracksail simulation did not cause this behaviour to be exhibited The angle at which the boat can sail to the wind is also different in each system hence the different angles on the upwind leg BeagleB was set to attempt to sail at 35 degrees to the wind while the HIL system requires at least 45 degrees due to limiations www A TIBOOK ir 122 C Sauz and M Neal in Tracksail The Tracksail simulation was actually set to 55 degrees due to code inherited from another robot which could only sail at 55 degrees The total time taken by the Tracksail simulation and the HIL simulation seem to differ massively this may be due to the way time is calculated The Tracksail simulation considers 1 second to have elapsed after each iteration of the loop while the HIL simulation uses a realtime clock to calculate this The number of sail move ments in the HIL simulation is also much higher than either of the other methods This is mainly due to a period when the sail
261. urned on and off almost instantaneously without a significant warm up time needing less than two seconds for spinning up This feature will further help to reduce power consumption on long term missions Already having an Ethernet network in place on the ASV Roboat leverages the integration of the radar into the control system It proofed of much help in the em pirical protocol analysis that the lower transmission layers comply to an industry standard 3 Structure of a LowRance Installation The Lowrance installation Figure 2 allows multiple sensors and consumers to be connected For example additional sensors can be transducers or NMEA heading sensors A Radar Interface Box is used to connect a power supply for the antenna unit and has a standard Ethernet port on the controller display side 4 Methods for Empirical Protocol Analysis All described methods and research had been done solely to establish compatibil ity between our autonomous surface vessel control system and the radar system mentioned above Neither the data nor the protocol incorporated any recognizable protection scheme To be able to record the data exchanged between antenna unit and controller unit we decided to use an Ethernet hub a switch with a mirror port would work as well HDS 7 HDS 10 aus Radar Expansion Interface Lowrance Port Box BR24 ee Fig 2 A typical installation as provided by the manufacturer 7 The optional Navico Ex pansion Port is
262. urrent data into the calculation of the travel time could also make the routing more accurate and may lead to better routes Using a graph where nodes are connected only to their eight direct neighbors puts a considerable restriction on possible paths and can lead to noticeable differences between projected and actual arrival time This restriction could be lowered while improving the quality of the calculated route by using a graph where nodes are con nected to 24 neighbors as presented in 3 We believe that this could be achieved with a reasonable increase in calculation time by replacing the CHR rules for node and edge generation References 1 Daniel K Nash A Koenig S Felner A Theta Any Angle Path Planning on Grids Journal of Artificial Intelligence Research 39 533 579 2010 2 Donnay J D H Spherical Trigonometry Interscience Publishers Hoboken 2007 3 Erckens H B sser G A Pradalier C Siegwart R Y Avalon Navigation Strategy and Trajectory Following Controller for an Autonomous Sailing Vessel IEEE Robotics amp Automation magazine 17 1 45 54 2010 4 Friihwirth T Constraint Handling Rules Cambridge University Press Cambridge 2009 5 Hart P Nilsson N Raphael B A Formal Basis for the Heuristic Determination of Minimum Cost Paths IEEE Transactions on Systems Science and Cybernetics 4 2 100 107 1968 6 Papadakis N A Perakis A N Deterministic Minimal Time Vessel Routing
263. use the 20 input frame By that the algorithm ignores waves and clutter interferes and decreases the rate of false alarms 112 Later on we define the Interesting Points concept The Interesting Points which survived the time correlation are eventually defined as targets Multiple targets can be found in each frame and each identified target is reported only once 3 The Algorithm 3 1 Initial Cleaning The marine environment suffers from variety of noises of different kind In the first step we apply a number of image blurring techniques to get a smoother texture of the sea The input image might contain many noised pixels that will interfere with reducing the search space This may become critical for real time performances Therefore a number of Gaussian softening filters are used In order to preserve the horizontal structure of the sea pattern we use convolution with a specific matrix designed to preserve the horizontal structure The initial cleaning part includes the Adaptive Smooth filter As well known in the image processing world this filter is designed to blur the image and save the sharp edges This characteristic becomes very crucial in later stages to recognize and identify the targets The Adaptive Smooth filter w x y is a very efficient filter but hard to deal with salt and pepper clutter The filter computes as _ BAG w x y e 2f 1 www A TIBOOK ir 130 O Gal where Ix 1 y I x 1 y 5 2 G x
264. veloped a systems architec ture that has proven robust and easily customizable suited to wide variety of tasks In this section we will outline the basic components of the system and dis cuss the unique aspects of the system design 2 1 Basic System Architecture As discussed previously the systems used for robotic sailing at USNA are devel oped with a functional block approach The primary subsystems and associated requirements are as follows where each subsystem is again managed as a func tional block that is under the purview of one subsystem manager who must com municate with the rest of the team to guarantee proper performance Processing This subsystem is the heart of the robot Here are carried out all of the computations required for planning and executing the assigned tasks SailBot re quires that all processing be on board and the form factor of the vessels designed precludes the use of a standard laptop or netbook Power issues and heat dissipa tion must also be considered in the selection process www A TIBOOK ir 96 B E Bishop et al Actuation Because the system is designed to allow a human to take over from shore or a chase boat at any time the actuators are selected to be R C friendly As of now only the main sail and the rudder are actuated although actuators have been selected for independent control of the jib as well The key concerns when choosing the actuators are power consumption accuracy and sp
265. very first tests of Breizh Spirit 1 were conducted during the WRSC 2009 in Porto Obviously they were not successful From discussions with the other r Ko Roscanvel We wu dur a D Vanv oc 4 Os d k b s Downolad from GPS 1 Fig 9 Brest Harbour crossing The yellow points correspond to the waypoints Breizh Spirit 1 sailed 6 5 nm at average speed of 3 knots www ATIBOOR ir Breizh Spirit a Reliable Boat for Crossing the Atlantic Ocean 67 participants it was clear that the development of such a project should lean on three tools e alight prototype easy to carry and implement e a simulator useful to detect bugs and to avoid spending time on water e a monitoring system to get information from the tests and understand the boat behaviour During the following weeks a monitoring system was implemented The first suc cessful tests were carried on the small lake Ty Colo near Brest Basically the boat managed to turn around 3 way points Then it was decided to attempt the crossing of the Brest Harbour from Cosquer to Lanveoc in early September The wind was north east 15 to 20 knots during the main part of the crossing and then dropped to less than ten knots and turned east to north east at the end The boat was totally autonomous until it reached the Lanveoc shore Then we took the control to end the course and to pass the Pen ar Vir point just in front of the French Naval academy Breizh Spirit 1 did 6 5 nau
266. w A TIBOOK ir Part V Localization and Route Planning www ATIBOOR ir www ATIBOOR ir A Digital Interface for Imagery and Control of a Navico Lowrance Broadband Radar Adrian Dabrowski Sebastian Busch and Roland Stelzer Abstract The paper describes a method to establish compatibility between an au tonomous surface vessel control system and a Navico Broadband Radar BR24 The solution obtains radar imagery and control of the antenna unit over its standard Eth ernet interface making the proprietary controller unit optional It presents devices software and methods used for empirical protocol analysis and documents the find ings Protocol details for the following functions have been identified Operation zoom level various filter settings scan speed and keep alive An open source imple mentation with basic operational functionality has been made available It features a live network mode and a replay mode using captured network traffic In live mode controlling radar operation as well as zoom level is possible In both modes the radar imagery stream is rendered and displayed 1 Introduction Radar is an important sensor in nautical navigation as it is also for autonomous robots Typically a radar systems vendor delivers an integrated solution A radar transceiver together with a display unit Radar provides perception of the Adrian Dabrowski Faculty member of H here Technische Bundeslehr und Versuchsanstalt Wien V HTL
267. ware Innovations 31 starting point _ sailboat Fig 3 We made SCILAB simulation We launched the boat in several different directions We observed that the boat is able to reach the desired course whatever the starting angle is which validates our approach simulation wise Consider 6 9 the angle of the wind vane in the global workspace o 9 7 64 0 7 As such F Frmax det uy u5 8 where F max 15 the force applied on the wind vane when it is perpendicular to the wind uy PT and ug cos 6 9 sin 6 9 2 2 2 Result of Stabilization Figure 3 represents the trajectories computed while starting the boat at different angles We observed that the boat is able to reach the desired course whatever the starting condition is 2 2 3 Navigation Using Only the Self Steering Device It is possible to navigate using only the self steering device by progressively chang ing the target angle Figure 4 shows the tacking and jibbing maneuvers for both case of the self steering device In fact the trajectories are the same However in reality the big sail will hide the wind vane of the self steering device in some of the cases As a consequence unless rising the wind vane above the big sail or the opposite the boat might have difficulties to jibe with a bow self steering device and to tack with a stern self steering device Note that it is necessary to have some speed to perform tacking or else the boat will be unable t
268. weden allowing for in excess of 500m line of sight distance 2 Clearly to get this range when communicating with an onshore computer a similar module and a high performance antenna must be used too Moreover as the dis tance between boat and onshore antenna increases the likelihood for transmission errors and re transmission is also rising This can cause additional latencies as il lustrated in Figure 4 While the average round trip time remains below 40ms for up to 300m distance the communication may also be affected by other boats e g when disrupting the line of sight Considering additional latencies due to the serial link message processing and the timer interval the total latency is typically below 180ms and could be further reduced by shortening the timer interval 3 3 Sensors Autonomous control requires reliable sensors which additionally need to be small lightweight and energy efficient to be used with the rrMM We equipped our boat with a number of relative and absolute sensors i e wind speed wind direction 3 axis compass 3 axis accelerometer 3 axis gyroscope and GPS For inertia measurements we used the combined magnetometer and accelerom eter 1M 3501 Amosense Korea and the gyroscope ITG 3200 Invensense CA Both sensors are interfaced using the I2C protocol The IM 3501 provides calibrated three axis magnetic field and three axis acceleration measurements at a frequency of www A TIBOOK ir A New Class
269. well as a USB to RS422 bridge to control the antenna unit This system is used by the SSC San Diego USV team 6 Within the project ROAZ II 1 a Furuno NavNet Vx2 radar based solution has been evaluated As a result the authors were unsatisfied that among other things the radar needs gain adjustments which they could only do manually For the ASV Roboat the team decided for a Navico Lowrance BR24 Broadband Radar 7 This radar uses Frequency Modulated Continuous Wave FMCW radar technology The decision was based on the low power consumption compared to l LSA Laborat srio de Sistemas Aut snomos Instituto Superior de Engenharia do Porto Austrian Society for Innovative Computer Sciences 3 Non Disclosure Agreement www A TIBOOK ir A Digital Interface for a Lowrance Broadband Radar 171 conventional pulse radars and the fact that Ethernet is used for communication in this system Pulse radars are typically rated in the kilowatt output power range whereas this one has a peak power output of 100 mW nominal and a total power usage of about 20 W at 12 V in operation Pulse radars measure the time of flight of a reflected elec tromagnetic signal In contrast FMCW radars send out signals of linearly increasing frequency and compare the frequency of an incoming reflection to the currently sent frequency Knowing the rate of frequency increase allows the calculation of dis tance Despite its low power consumption FMCW radars can be t
270. which had occurred in low levels of surf onto a mixed pebble and sand beach suggested that attempting to autonomously beach a MOOP will almost always be damaging to the hull and in more severe situations would probably result in serious damage The swaying caused by the wind sensor is a serious issue and suggests that the wind sensor weighed too much approximately 300 g therefore it will not be possible to use a Furuno rowind sensor on a MOOP 3 5 WRSC 2010 MOOPI and MOOP3 were entered into the 2010 World Robotic Sailing Champi onships in Kingston Ontario Canada They both struggled against a current of ap proximately one knot crossing the racing area Both boats were able to successfully sail with the current but never completed an entire course due to the current It was interesting to note that the 53 cm MicroMagic boats from the University of L beck were able to complete the same course This suggests it is specific design el ements of the MOOP rather than simply the lack of hull speed in any small boat causing these problems These boats featured much narrower keel designs than the MOOPs and therefore are likely to have encountered far less drag than a MOOP hull www A TIBOOK ir 52 C Sauz and M Neal 4 Conclusions and Future Work Overall the MOOPs have proven themselves to be a cheap flexible and reliable plat form for developing sailing robot control systems However they have struggled with sailing upwind and against cu
271. wing calculation w_dir_value x 1025 1 360 w_dir_period 1024 Where w_dir is the wind direction in degree w_dir_value the active phase of the PWM signal and w_dir_period the whole period of the PWM signal wdir 1 3 3 Actuators To control rudder and sail we use standard components similar to the components used for the Saudade a servo for the rudder and a sail winch with separate power supply to control the sails Graupner Germany Figure 5 shows the components of the test system 3 4 Power Consumption For a small boat like our test system Saudade the power consumption is a critical factor To keep the system simple and cheap we searched for a control system which can be supplied by small NiMH oder LiPo Lithium polymer accumulators The chosen LM3S6965 needs less than 1 W The power consumption and the cost of the components shows Table 3 Even if the computing power may not be sufficient for all algorithms we think this is a good starting point European Geostationary Navigation Overlay Service www A TIBOOK ir Using ARM7 and uC OS II to Control an Autonomous Sailboat 107 Table 3 Power consumption and cost of the control system components Component Current at Voltage Cost EKS LM3S6965 at 50 MHz display on 170 mA at 5V 80 Bluetooth modem 30mA at 5V 54 GPS 80 mA at 5 V 50 Compass 5mA at 3 3 V 130 Wind direction 20mA 3 3 V or5V Rudder servo quiescent current 10mA at 5V 50 Rudder servo max
272. with communication gateway CG and multiple clients The CG handles the Bluetooth connections to the boats and directly forwards all data to the World Server The server also receives data from clients not using the CG and offers all gathered information to all clients The graphical user interface GUI and the not yet implemented judge connect to the World Server as an ordinary clients too e g their position heading or speed This data also includes bouys and obstacles as required and can be polled by the clients Every object has a unique ID and a position specified by latitude and longitude GPS coordinates The individual objects also carry more specific information like a radius for buoys and circular obstacles or wind direction and wind speed for sailboats Our world server stores all received information in a log file for later analysis but only keeps the most recent data for each object in the main memory and available for clients If a client application needs a history of certain events for example for some kind of prediction algorithm it is up to the application itself to maintain such a history This reduces the complexity of the data management as well as the API of the world server to a minimum The API consists mainly of only two methods one that retrieves a collection of all objects currently registered at the server with their up to date data and another method that allows the clients to add or update their own object
273. x being the state of the system For the simulation we use Euler approximation to estimate the state of the robot in real time as follows X41 f x u4 dt x 2 x being the state of the system and u being the command at time step k For the simulation of the robot we used the state equations from 5 Figure 2 shows the block diagram of the system boat wind vane regulator Denote by and respectively the angle between the wind vane and the axis of the boat and the angle between the rudder and the axis of the boat as shown in Figure 1 5 9 is the sail command which will be left constant for the simulation In fact our prototype have an non actuated sail to add robustness Denote by V the wind vector Denote by the angle between the rudder and the wind vane when the www A TIBOOK ir 30 J Sliwka et al wind V boat orientation speed 9 Vv D ig 5 gt sail gt Bi Regulator rudder Boat rs force of the wind on the regulator Fig 2 This figure represents the block diagram corresponding to the system boat wind vane regulator The regulator will control the angle of the rudder of the boat and exerts a force F on the boat due to the drag of the wind vane The regulator will be moved by the wind V The angle between the rudder and the wind vane when the rudder is in zero position is determined by the variable The sail command is left constant rudder is in zero
274. y cost around 500 parts only 2010 prices are 72 cm long weigh approximately 4 kg and follow a long keel design inspired by a Nordic folk boat Their small size is intended to make them easy and cheap to produce in large quantities It also enables them to be easily handled by one person to not cause problems for other marine traffic to be transported in a small car or as baggage on an airliner for use abroad Their low cost enables them to be deployed in large quantities to be lent to students to be adapted Colin Sauz Mark Neal Department of Computer Science Aberystwyth University e mail cos mjn aber ac uk www A TIBOOK ir AO C Sauz and M Neal for experimental purposes or to be sent on high risk missions with poor chances of recovery 2 Boat Designs A number of design goals were considered when the MOOPs were initially pro posed Key to these was to produce a small cheap robust and lightweight boat to cross the Atlantic Ocean as part of the Microtransat Challenge It was suggested that several small cheap and simple boats might between them stand a greater chance of one succeeding than a single larger more expensive boat A long keel design was selected in order to have the rudder extend from the back of the keel and reduce the possibility of seaweed and other debris from becoming entangled in it To avoid making any holes in the hull itself a magnetic linkage was used between the rudder actuator and the rudder
275. y in the process This test took place during calm weather Beaufort force 1 or 2 and so cannot be considered to have covered conditions representative of those on the high seas 3 4 Aberystwyth Sea Test MOOP2 had been constructed with the intention of cross the Atlantic ocean during the later cancelled 2009 Microtransat Challenge A short sea trial was undertaken in August 2009 in the waters off Aberystwyth A waypoint was set approximately 200 m offshore and a second waypoint was set onshore with the intention of having the boat sail itself ashore in order to test the concept of setting the final waypoint ashore to have the robot beach itself The robot was sailed some distance towards its first waypoint but encountered some difficulties On several occasions the robot span around in circles and due to the weight of the wind sensor it constantly swayed from side to side after each wave passed Eventually the chase boat decided that there may have been some software issues and to abort the attempt and advance the boat to its next waypoint The robot was able to successfully sail back towards the beach sailed through the relatively small 0 5 m surf and onto the beach This process was repeated three times to test repeatability and if the boat would actually become beached or if it would be swept away by the next wave After the third attempt several dents and holes were noticed in the hull and further tests were aborted The level of damage
276. y the Microtransat idea of Yves Briere ISAE France and Mark Neal Aberystwyth University Wales UK 15 14 The organizers describe the Microtransat in on their web site 29 as follows The Microtransat Challenge is a transatlantic race of fully autonomous sailing boats The race aims to stimulate the development of autonomous sailing boats through friendly competition Participating sailing boats shall further be small max 4 m in length unmanned and use wind as the only form of propulsion A few smaller Microtransat competitions took place prior to the real transatlantic race These allowed contestants to exchange ideas and test their boats in less harsh environments The first transatlantic attempt was in 2010 The only boat which entered the com petition was Pinta from Aberystwyth University Wales UK According to Colin Sauz the boat was 49 h and 87 km under autonomous control before the computer system failed 3 2 2 SailBot SailBot is an international competition for autonomously controlled sailboats Aimed primarily at undergraduate student teams the goal is to give engineering students a practical application of the topics they have learned while also providing a fun way to learn project management in a multidisciplinary environment A successful Sail Bot balances the needs of naval architecture mechanical engineering systems and electrical engineering as well as project management 9 SailBot competitions were held
277. ying operating and recovering a real robot These overheads can significantly slow development of control systems hardware and software and can create difficulties in comprehensively testing all aspects of a control system A simulator offers a mechanism to overcome these limitations and produce better tested control systems However simulators also suffer from a number of key disadvantages it is difficult to create a realistic model of a boat s motion through water especially one which can be executed in real time and to model the detailed effects of waves tides winds shifts and gusts Colin Sauz Mark Neal Department of Computer Science Aberystwyth University e mail cos mjn aber ac uk www A TIBOOK ir 114 C Sauz and M Neal This paper outlines our experiences in simulating sailing robots It focuses on efforts to produce software only and hardware in the loop HIL simulators These results are compared against sailing similar voyages on real robots 2 Background In previous work 9 a simulator based upon an open source sailing game called Tracksail 6 An API was implemented by sending commands over a TCP IP socket between a client application and Tracksail This interface followed a mechanism similar to that which we used in some of our early sailing robots This is based on a highly simplistic physics model and makes no attempt to simulate the actions of waves currents or tides It allows variable wind directions th
278. ystwyth Univer sity 12 most boats have been one offs This is reflected by the rather coarse design rules implied by two existing classes the Microtransat and SailBot class One ad vantage is the room to try really innovative hull and rig designs Yet a disadvantage lies in the fact that boat performance is hard to compare and any rating formula is Alexander Schlaefer Daniel Beckmann Maximilian Heinig Ralf Bruder Institute for Robotics and Cognitive Systems University of Luebeck Ratzeburger Allee 160 D 23562 Luebeck e mail schlaefer heinig bruder rob uni luebeck de daniel beckmann web de www A TIBOOK ir 22 A Schlaefer et al typically seen as either unfair or too complicated or often both Hence introducing a more restricted class allowing for direct comparison of boats and methods would be interesting Few of the existing designs would be practical as a prototype for such a new class The MOOPs are built for long term autonomy and durability but lack sailing performance particularly upwind A number of SailBot boats have beend designed at the United States Naval Academy 8 9 While the sailing performance is excel lent they are currently custom built and not widely available Moreover the length and draft 2 0m and 1 5m respectively make the boats harder to launch This is even more problematic with the larger Microtransat boats e g the Welsh Beagle B 11 the Swiss Avalon 5 the Portuguese FASt 1

Download Pdf Manuals

Related Search

Related Contents

Opal Operating Instructions [hires]    Erste Hilfe für Ihr Notebook - *ISBN 3-8272-6978-4* - © 2005  ES1224 User`s Manual  Bedienungsanleitung/User manual  Metapace T-3  Artwizz ScratchStopper Matt  Denon DN-C110P  Guia do Usuário do Altiris™ Asset Management Suite 7.1 da  Manual do Proprietario_Lisboa.indd  

Copyright © All rights reserved. Failed to retrieve file