an embedded system for small
Contents
1. 43 11 1975 1983 Jung D Levy E J Zhou D Fink R Moshe J Earl A and Tsiotras P 2005 Design and development of a low cost test bed for undergraduate education in UAVs In Proc of the 44th IEEE Conf on Decision and Control and the European Control Conference 2005 Jung D and Tsiotras P 2007 Inertial attitude and posi tion reference system development for a small UAV In AJAA Infotech at Aerospace Kaminer I Yakimenko O Dobrokhodov V Lizaraga M and Pascoal A 2004 Cooperative control of small uavs for naval applications In Proc of the 43rd IEEE Conf on Decision and Control Kogan D and Murray R M 2006 Optimization based navigation for the DARPA Grand Challenge In Proc of the 45th IEEE Conf on Decision and Control Lee S Lee T Park S and Kee C 2003 Flight test re sults of UAV automatic control using a single antenna GPS receiver In AJAA Guidance Navigation and Control Conference and Exhibit Mettler B 2003 Identification Modeling and Character istics of Miniature Rotorcraft Boston Kluwer Aca demic Publishers Mettler B Tischler M B and Kanade T 2000 System identification of a model scale helicopter Technical Report CMU RI TR 00 03 Carnegie Mellon Univer sity Mettler B Tischler M B and Kanade T 2001 System identification modeling of a small scale unmanned he licopter Journal of the American Helicopter Society M2. down datalink A convenient Head Up display system on the ground station was designed for that purpose Our data link is a MaxStream module operating at 2 4 GHz It provides a RS232 serial data link at 9600 baud send ing the information from the embedded system to the ground station 3 2 Small scaled Helicopter We use a Benzin acrobatic helicopter from Vario with Graupner electronics C4441 servomotors which have high speed and high torque 8 5 Nm a 16 channels receiver and a yaw hobby gyroscope It is a very reliable helicopter platform we never observed any mechanical issue over mode than 100 flights Its payload capacity 4Kg is large enough for numerous reconnaissance applications Its rotor is 1 9 m wide 3 3 Interfacing and Vibrational Issues Wiring the embedded system to the existing heli copter circuitry was achieved using some specific ad ditional boards and connectors To measure the pilot s orders in real time we used a 6 channels voltage fol lower circuit Numerous LEDs were added to check the status of our system A central problem observed on board helicopters is the 25 Hz vibrations induced by the main rotor blades These vibrations generate a large amount of noise on the inertial sensors In practice these noises totally overwhelm the useful signals Fortunately it is possible to solve this issue by using well chosen noise dampers On our helicopter we decided that the micro controller and the s
3. obstacle avoidance and navigation we will connect LADARS and ultrasonic sensors to our embedded system ACKNOWLEDGEMENTS The authors are indebted to the numerous students technicians and engineers who have been collaborat ing to the development of the presented technology and have brought their support during the conducted experiments REFERENCES Caccamo M Baker T Burns A Buttazzo G and Sha L 2005 Real time scheduling for embedded sys tems In Hristu Varsakelis D and Levine W S ed itors Handbook of networked and embedded control systems Birkh user Castillo P Lozano R and Dzul A E 2005 Modelling and control of mini flying machines Advances in in dustrial control Springer Cheng J Lu Y Thomas E R and Farrell J A 2006 Data fusion via Kalman filter GPS and INS In Ge S S and Lewis F L editors Autonomous mobile robots Control engineering series Taylor and Francis Cloud Cap Technologies 2004 http www cloudcaptech com Hood River OR USA Cremean L B Foote T B Gillula J H Hines G H Kogan D Kriechbaum K L Lamb J C Leibs J Lindzey L Rasmussen C E Stewart A D Bur dick J W and Murray R M 2005 Alice An information rich autonomous vehicle for high speed desert navigation Journal of Field Robotics 9 Hamel T and Mahony R 2007 Image based visual servo control for a class of aerial robotic systems
4. AN EMBEDDED SYSTEM FOR SMALL SCALED AUTONOMOUS VEHICLES David Vissi re and Nicolas Petit D l gation G n rale pour l Armement France Ecole Nationale Sup rieure des Mines de Paris France david vissiere dga defense gouv fr nicolas petit cas ensmp fr Keywords Abstract Embedded systems autonomous vehicles UAVs We consider the problem of designing a modular real time embedded system for control applications with unmanned vehicles We propose a simple and low cost solution Its computational power can be easily improved depending on application requirements To illustrate its performance we report the implementation of a 75 Hz Extended Kalman Filter used for state estimation on a small scaled helicopter 1 INTRODUCTION In this paper we present some of our research effort in a current program aiming at proposing control strate gies and a control architecture for a group of hetero geneous autonomous vehicles To conduct this research we designed a versatile and simple real time embedded system which can be easily used as real time guidance and navigation sys tem on various platforms Our work focuses on het erogenous vehicles including small scaled typically less than 2 m wide Vertical Take Off and Landing aerial vehicles VTOL as in Castillo et al 2005 or fixed wing aircraft and ground vehicles with tank like dynamics as in Morin and Samson 2006 Vissi re et al 2007 In the future these v
5. among the family of 32 bits kits the MPC555 has substantial computational capabili ties and a large number of versatile and programmable Input Output ports In particular we make an exten sive use of its TPUs Time Process Units UARTs A D converters SPIs MPIOs Modular I O system see Motorola 2000 Finally it is small credit card format and has a low weight e Magnetometer We use a HMR2300 three axis magnetometer from Honeywell Its range is 2 gauss and it has a 70 pgauss resolution No operating system is used on the micro controller Rather the MPC555 runs a specific interrupts driven software written in C Information 159 ICINCO 2008 International Conference on Informatics in Control Automation and Robotics Vehicle Sensors actuators td J Computation P Power PC S Board Ground Remote station Control Figure 2 Sensors and computation board connections to the central micro controller 412V i Computation Board IMU ascent TPU gt 4DATALINK gt GPS main c kav BAROMETER Ground E anak Station 4 Remote L COMPASS QQPWH DASA Control L veHicLe ee nterrupt t actuators h SWITCH Receiver 5V Figure 3 Embedded system internal connections from each sensors is transfe
6. asurement Unit IMU Our IMU is a 3DM GX1 from Microstrain It contains three angular rate gyroscopes three orthogonal single axis magnetometers and three single axis accelerometers along with 16 bits A D convert ers and a micro controller This IMU can deliver different messages ranging from raw data to rec onciliated measurements In our setup we ask the IMU to deliver only calibrated sensors data at a 75Hz rate e Global Positioning System GPS Our GPS is a TIM LS from pblox Through a proprietary binary protocol it provides position and velocity information at a 4Hz rate Position error is 2 5 m Circular Error Probability CEP and velocity er ror is 2 m CEP The GPS receiver is not very tolerant against power supply voltage ripples These can be kept below the 50 mV requirements thanks to a dedi cated power supply regulator from TRACO e Barometer Our barometer is the MS 5534 from Intersema Using a SPI type protocol it gives calibrated digital pressure and temperature infor mation This device requires a 3 V power supply which is obtained through a fast response diode from the main 5 V power supply of the micro controller e Anemometer A PXLA sensor from ASensTec delivers a differential pressure analog signal which can be read through a 10 bit A D converter 7530 20 e Take off and Landing Detector Being able to detect take off and landing instants is necessary to properly ini
7. cessor is a 1 2Ghz C7 M from VIA designed for embedded applications It can perform 1500 MIPS and has clas sic PC Input Output ports such as a UART serial port used as main data link with the micro controller an ethernet board not used here a VGA screen output which can be used to monitor the system during de bugging phases of the software and hardware devel opment a keyboard and 4 USB ports which can be considered for plugging future devices such has con trollable cameras Experimental preliminary tests have shown us that multi threading one thread for message decoding and one thread for the main algorithm presents two major drawbacks some data can be lost and the calculation cycle may end unexpectedly slightly late For this reason we decided to write our own UART driver us ing an interrupt handler in the kernel space Further we de activated all hardware interruptions associated to unnecessary devices As a result only interrupts from the UART are enabled Finally we used a single computation thread The operating system is installed on a bootable 1 Gbyte Disk On Chip system which prevents all possible mechanical failure associated to hard drives This flash memory device is directly connected to the IDE port of the mother board The board is powered by a pico PSU power supply which provides various voltages ranging from 5V to 18V The computation software are written in C and can be either be updated directl
8. e actuators transfer functions and the yaw rate gyroscope In some works see Mettler 2003 or Mettler et al 2000 actuators servomotors are considered as first order systems with dead band We identified such transfer functions for various Graupner and Futaba servomotors Results of various tip in and tip out in reference signals were recorded to compute the time constant of the first order model On board our helicopter a hobby gyroscope from 161 ICINCO 2008 International Conference on Informatics in Control Automation and Robotics g g 20 30 40 50 60 70 Figure 5 Bode diagram showing the resonance peak and the cut off frequency of the mechanical structure equiped with the sensors the micro controller and the spring dampers suspension The various plots are obtained on varying locations on the vibrating structure and show a good spatial uniformity of the vibration damping Graupner is used to help the human pilot keeping the yaw rate as small as possible Pilot orders are transferred from the R C receiver to the tail actuator through this gyroscope To validate simple models of this transfer we put our IMU under this gyroscope to measure the angu lar velocity Simultaneously we connected the gyro scope and recorded the gyroscope signals sent to the tail actuator Surprisingly it was discovered that the gyroscope feedback behaves as a 2 Hz low pass filter on the pilot orders and directl
9. e presented In Cremean et al 2005 high speed data fusion systems have been de veloped in view of the DARPA Grand Challenge In this later experiment several technological breakthroughs are presented using a high end and powerful computer architecture Software compo nents communicate in a machine independent fashion through a module management system Our experiments can not use such a high end setup because the typical payload of our aerial ve hicles does not exceed 5 kg Much smaller and lower weighting systems can be considered though In Jung and Tsiotras 2007 an embedded system is proposed which does not incorporate any powerful calculation board A simple Rabbit Semiconductor RCM 3400 micro controller is used to perform complementary filtering data fusion using a limited computational power In the same spirit in Jung et al 2005 a low cost test bed for UAVs is presented It is reported that the main advantage of designing such an autopilot from 157 ICINCO 2008 International Conference on Informatics in Control Automation and Robotics Figure 1 Cooperative autonomous vehicles in a future bat tlefield scratch is that by contrast to commercially available products Cloud Cap Technologies 2004 Micro Pi lot 2004 it provides full access to the internal con trol structures We totally agree with this point of view In this paper we present a solution lying in the middle of the two previous
10. ehicles will be asked to act cooperatively on the battlefield as pic tured in Figure see also Kaminer et al 2004 Olfati Saber 2006 for other scenarios In practice the aerial vehicles represent the most challenging applications in terms of navigation and guidance The main reason for this is that these ve hicles can not easily go into any safe mode as op posed to the ground vehicles which are in compar ison slower and simpler While it was proven that with lowered performance expectations it is possi ble to stabilize a fixed wing Unmanned Air Vehicles UAV by directly closing the loop with signals from well chosen sensors e g in Lee et al 2003 the authors propose a solution to automatically control a fixed wing UAV using only a single antenna GPS re ceiver it is considered by the vast majority of the UAV community that navigation systems require data fusion Cheng et al 2006 In facts each sensor technology has its own flaws among which are drift noises and possibly low resolution or low update fre quency Yet large factors of accuracy can be gained by reconciliating their data Example of on board data fusion applications are ubiquitous among autonomous vehicle control experi ments Reconciliating GPS and Inertial Measurement Unit IMU measurements is a classic case study In Xiaokui and Jianping 2002 results of data fusion from a BeeLine GPS receiver from Novatel and a miniaturized IMU ar
11. ensors would all be lo cated on a board which would be physically con nected to the frame of the helicopter through four Figure 4 Our embedded system fitted into the custom built landing gear of a small scaled Vario Benzin heli copter Springs and dampers are used to filter out vibrations from the main rotor blades spring damper systems see Figure 4 Experiments conducted on a vibrating table have shown that it was advantageous to keep the embedded system as com pact and as rigid as possible The total weight of the subsystem is about 600 g We decided to attach some of the batteries to it to bring the weight close to 1 8 kg This enabled us to use off the shelf dampers yield ing appropriate cut off frequencies MV801 5CC dampers from Paulstra were chosen for their abil ity to work with low masses vibrating at low frequen cies With these we obtained a satisfactory vibration damping with a cut off frequency around 9 Hz This is represented in Figure 5 Further resonant frequen cies due to the engine frequency around 160 Hz the tail rotor frequency around 115 Hz and the tail boom were removed using a digital notch filter The presented solution attenuates high frequency vibra tion inputs down to negligible levels 3 4 Experimental Identification Preliminary model identification experiments need to be conducted before state estimation can be per formed In particular using our embedded system we studied th
12. icro Pilot 2004 http www micropilot com Stony Mountain Canada Morin P and Samson C 2006 Trajectory tracking for nonholonomic vehicles In Kozlowski K ed itor Robot Motion Control Recent Developments Springer Motorola 2000 MPC555 MPC556 user s manual User s manual Motorola Murray R M Hauser J Jadbabaie A Milam M B Pe tit N Dunbar W B and Franz R 2003 Online control customization via optimization based control In Samad T and Balas G editors Software Enabled Control Information technology for dynamical sys tems pages 149 174 Wiley Interscience 163 ICINCO 2008 International Conference on Informatics in Control Automation and Robotics Olfati Saber R 2006 Flocking for multi agent dynamic systems Algorithms and theory JEEE Trans Au tomat Control 51 3 401 420 Simon D 2005 Optimal state estimation Wiley Vissi re D Chang D E and Petit N 2007 Experi ments of trajectory generation and obstacle avoidance for a UGV In Proc of the 2007 American Control Conference Xiaokui Y and Jianping Y 2002 Study on low cost GPS DMU integrated navigation system In AIAA AAS Astrodynamics Specialist Conference and Exhibit 164
13. lculation cycles to simultaneously read or send data and perform computations Some ex perimental results are presented in Figure 7 Position estimates around hovering are reported In practice they appear to be in great accordance with recorded videos 4 CONCLUSIONS Designing an embedded system which can qualify as a control system for various small scaled air and ground vehicles is the subject of the research project presented in this paper The embedded system we propose here has some interesting features It is sim ple low cost and most of all easy to upgrade Its two processors architecture can incorporate various new AN EMBEDDED SYSTEM FOR SMALL SCALED AUTONOMOUS VEHICLES Figure 7 Position estimates during a hovering flight processors and sensors In our laboratory this embedded system has been successfully used on a fixed wing aircraft Ras cal 110 a small scaled helicopter Vario Benzin trainer and ground vehicles Pioneer 4 from Mobile Robots Further developments focus on giving more au tonomy including path planning algorithm follow ing Kogan and Murray 2006 Murray et al 2003 to respond to high levels orders from a remote user We will surely need more computational power which can be obtained by simply upgrading the computa tional board In parallel we develop a new aerial ve hicle for urban area applications For this last project we consider using other sensors In particular for
14. ly mentioned categories Our system uses two processors One processor is used to gather data from the sensors and to control the actuators The other processor is used to perform the data fusion calculations and possibly the control algorithms The advantages of this structure are as follows i task scheduling is easily programmed be cause only one of the two processors is in charge of handling the numerous devices and I O ii the com putations are performed as one single thread on a ded icated board PC type iii depending on the com putational requirements the computation board can be easily upgraded without requiring any software changes or rising any concern about task scheduling iv finally the overall system is quite low cost since it relies on off the shelf components and can be easily maintained The paper is organized as follows In Section 2 we present our system architecture We detail our hardware components and comment on their choices In Section 3 we present as a test case the embedding of our system into a small scaled helicopter Numer ous details of implementation are provided Finally we conclude and give directions of future work 2 SYSTEM ARCHITECTURE Our primary goal was to develop an embedded system to test algorithms of various complexity on board var ious small scaled platforms Early in the design pro cess one first constraint which appeared to us was the payload limitations of the considered f
15. lying machines This lead us to focus on designing a low weight em bedded system 158 A second issue that was also raised early in the design stage was that the real time requirements of a control system for such small UAVs are very strong This is mostly due to the short time horizons instabil ities Yet in the context of embedded systems real time scheduling of a number of sensing and compu tation tasks is known to be a difficult problem More precisely as exposed in Caccamo et al 2005 the problem of determining the feasibility of a periodic sequence of prioritized tasks is often NP hard Suf ficient but not necessary tests are pessimistic Pop ular strategies such as the Rate Monotonic policy see again Caccamo et al 2005 which consists of putting the highest priority on the shortest task can be proven to be unfeasible is the CPU load is too large While being troublesome on ground vehicles such in feasibilities and the induced inconsistencies in the embedded calculations would represent a cause of potential major failure for our aerial platforms Keeping these two considerations in mind we de cided to develop a robust two processors embedded system running two distinct softwares and communi cating through a simple two ways protocol The sys tem specifications are as follows 1 It performs the sensing and calculation tasks sep arately 2 It is fast enough to run a typical 15 to 30 dimen sional sta
16. rred using a dedicated in terrupt handler routine Each external source or group of sources has its own interrupt level which avoids po tential conflicts Each data link is associated with a checksum to validate reception The data acquisition software running on the micro controller is event driven by the IMU messages which periodically sends 31 bytes of data Once the message of the IMU is received and validated by the micro controller others sensors information are ei ther directly read or picked in data buffers which are constantly fed with serial messages from the sensors through hardware interrupts Information is gathered in a 116 bytes message containing all the onboard measurements This message is sent to the calculation board through a high speed serial port Once the mes sage is received and validated the calculation board carries out one navigation loop consisting of a predic tion equation and an estimation equation of a Kalman filter 2 3 PC Computation Board The computing board is a PC running the Knop pix 3 8 1 Linux distribution The PC board was se lected among numerous models mostly mini ITX and PC104 based on computational power energy consumption toughness and price A fan less board was considered as the most relevant choice due to the often observed failures of fans in mechanically dis turbed environments 160 The chosen fan less calculation board has a stan dard mini ITX PC architecture Its pro
17. tes EKF algorithm with a low latency to eventually produce satisfactory closed loop re sults 3 Itis easy to upgrade 4 It is versatile enough to handle various type of sensors and communication protocols As exposed in Figure 2 and Figure 3 this mod ular embedded system is composed of a micro controller which is in charge of gathering information from all the sensors and a calculation board These two elements are connected by a serial interface The micro controller also has a downlink to a ground sta tion We now present the details of the hardware com ponents of our system 2 1 Sensors Considering both ground and aerial vehicles control applications we listed a series of useful sensors that needed to be incorporated into our embedded system Among these are an IMU a GPS receiver a pres sure sensor an anemometer magnetometers and var ious switches Other possibilities include odometers the upper limit on admissible load is 69 approxi mately AN EMBEDDED SYSTEM FOR SMALL SCALED AUTONOMOUS VEHICLES LADARs as used in Cremean et al 2005 and sonars as used in Vissi re et al 2007 or cameras as used in Hamel and Mahony 2007 In the con text of our study we only considered low cost sen sors We now detail these In each case we specify the weight in g the cost in USD the dimensions in mm the update rate f in Hz and the protocol of communication Comm e Inertial Me
18. tialize data fusion algorithms In de tails detection of the corresponding switches in the dynamics defines when the controls actually have an effect on the system This is not the case when an UAV is on the ground This detection is performed with on off switches which can be located e g on the landing gear They deliver a logic signal which can be readily interpreted To prevent electric arcs which might cause trouble to the connected micro controller we added a spe cific interfacing circuit This switches can also be replaced by active switches which can be used to activate various devices such as digital cameras or parachutes C0 6 25105 75 Boolean Our system is data driven by the IMU The main reason behind this choice is that the IMU is consid ered as a critical sensor 2 2 MPC555 Micro controller The micro controller which serves as an interface for the sensors and actuators is a MPC555 Power PC from Motorola It runs a specific software we developed using the Phytec development kit The reason for choosing this micro controller are as follows This device provides a double precision floating point unit 64 bit which is convenient for po tential embedded algorithms even if we do not use this possibility here since all computations are per formed on the calculation board it has a relatively fast 40 MHz clock it has 32 bit architecture and 448Kbyte of Flash memory and 4 MBytes of RAM Most importantly
19. y on the board via a ssh connection or transferred in a compiled form from a remote PC Custom scripts for compiling and distributing our executable code and configuration files are an efficient way to upgrade the software during development and testing 170 180 40 3 FIELD EXPERIMENT We now report some field experiment using our em bedded system To perform model identification tasks and design a real time state observer in view of closed loop control we decided to load our embed ded system into a small scaled helicopter In this sec tion we expose this work and give numerous details about solutions to specific interfacing and vibrational issues AN EMBEDDED SYSTEM FOR SMALL SCALED AUTONOMOUS VEHICLES 3 1 Experimental Setup We conducted all the experiments with the help of a human pilot which could at any time remotely con nect the inputs of the actuators of the helicopter to the outputs of our embedded system or directly to the out puts of the radio receiver and thus take direct control of the helicopter This is achieved thanks to a re motely controlled switch board which we designed This switch is actuated from our Graupner MC24 remote controller using one of the 12 available radio channels This system provides a rapid transition be tween manual and autonomous flight During all flights the pilot could see images from an on board camera and read measurements from the embedded system received for the
20. y transmits the opposite of the received angular rate without filtering it This can be summarized under the form Spilot Seyro Nem gyro r m 1 Tgyros 3 5 State Estimation The helicopter is a 6 degrees of freedom mechanical system with high bandwidth dynamic Reconstruct ing the full state of this system from low cost sensors only is quite a challenge We use an Extended Kalman Filter EKF see e g Simon 2005 to estimate the state of this system In practice the state of our EKF is composed of 23 variables including configuration states 13 variables using quaternions instead of Euler angles and the aerodynamic and external forces and torques 10 variables including the harmonic expan sion of the flapping phenomena as exposed in Mettler et al 2001 We used equally valued tuning parame ters for the 3 axis These are chosen to capture fast dynamics statistics data are Oacceleration 8 m s 2 162 D Receive data J State Estimation 3 Send Data a Figure 6 Succession of steps of data transfer and computa tion and Grorque 4rad s respectively Classically discrete time updates are implemented As already discussed updates are synchronized with the 75Hz measurements from the IMU The succession of tasks performed by the compu tation board of our embedded system is described in Figure 6 As can be seen it is necessary due to a re ception time being superior to the available time be tween two ca
Download Pdf Manuals
Related Search