Home

View/Open

1. lage en simulator for Chirp roboten Den skal v re kapabel til a simulere flere hundre roboter samtidig i sanntid Det er viktig at simulatoren er realistisk nok til a kunne brukes til prototype eksperimenter som til slutt skal kunne kjgre uten a matte endre grunnprinsippene bak ideen samtidig som den er enkel nok til a kunne handtere alle objektene i den simulerte verden For a oppna disse malene ble flere forskjellige metoder testet mot hverandre for hvert subsystem i simulatoren Fors k har vist at simulatoren klarte kj re et par hundre roboter samtidig med beregningstid til overs hvert steg i simulasjonen P grunn av tidsbegrensninger i prosjektet er deler av simulasjonen for lite realistisk f eks friksjonsmodellen til objekter som kan dyttes Dette burde kunne forbedres relativt lett Simulatoren bruker et relativt enkelt nettverksgrensesnitt for la kontrollere kj re helt separat av simulasjonen Kontrollere kan kun kommunisere med simulatoren ved bruke en UDP socket til styre en robot Dette lar brukere komme igang med en kontroller relativt raskt 111 Preface This report has been written by Robert Versvik as part of a master s thesis at the Norwegian University of Science and Technology NTNU during the spring of 2013 The master s thesis concludes my five year integrated study at the Department of Computer and Information Science IDI The purpose of the project is to build a simulator for the Chirp robot capab
2. My 10 I 24 APPENDICES Attributes string type must be Wall not case sensitive object position has two attributes float x x component of the position float y y component of the position float angle start rotation in degrees list hulls list containing hull objects which are lists of position objects object element position object containing an x and y coordinate float x x component of the position float y y component of the position 3 7 4 PushableObject Object The PushableObject object is similar to the Wall object only implemented as a dynamic body instead of a static body This means that robots and other objects can push them around unlike the walls that are immovable It currently uses dampening to emulate a bit of friction but it should be improved to properly simulate the friction instead The difference between the Wall object and the PushableObject object is the density attribute This is used by Box2D to calculate the mass of the object Listing 3 4 PushableObject Object type PushableObject position N DO peas 10 3 density 1 0 angle 180 Bahn 3 7 WORLD FORMAT 25 i 20 300 y SAG IP E Mza Op Vyle OM JE af gle De yde OM AE SU Do PUE 10 x 10 y 10 Attributes string type must be PushableObject not case sensitive object position has two attributes float x x component of the position float y y component of the position fl
3. NTNU Trondheim Norwegian University of Science and Technology ChirpSim A Prototype of a Simulator for the Chirp Swarm Robot Robert Versvik Master of Science in Computer Science Submission date June 2013 Supervisor Pauline Haddow IDI Norwegian University of Science and Technology Department of Computer and Information Science Abstract This project attempts to create a simulator for the Chirp robot capable of running a large amount of robots concurrently in real time It also needs to be realistic enough to be used for prototyping experiments meant to run on the physical robots while being simple enough to be able to handle the amount of objects in the simulated world To achieve these goals different approaches for each subsystem of the simulator have been evaluated against each other Experiments showed that the simulator was able to run a couple of hundred robots simultaneously while still having some computational time to spare each frame Due to time constraints it lacks some realism in how certain objects behave like friction on movable objects It is possible to improve the realism of the simulation relatively easily A simple networking interface was added to allow controllers to run completely separate of the simulation communicating through UDP to control a robot This allows users to quickly get a controller interfacing with the simulator to start experimenting ii Sammendrag Dette prosjektet pr ver
4. This can be used to implement a fake communications channel simulating how the real Chirp robots communicate over the distance sensors A controller message looking something like lt id gt sendmessage lt message gt and lt id gt getmessages could be used as an API to support message sending Alternatively the robot controller can receive the messages through a push func tionality where the simulator receives a message to be sent out on a specific sensor and automatically sends it to the robot being hit by the sensor 3 2 FUTURE WORK 13 Messages should be in the same format as the physical robot uses which has yet to be defined at the time of writing 3 2 2 World Editor The simulator as of now has no way to visually edit a world relying on users to manually edit the JSON file detailing the world instead An editor for simulation worlds would be a very beneficial addition to the simulator enabling users to visually edit their worlds to suit their experiments I think the editor should be a separate program so to not affect the simulator itself and avoiding complexity in the user interface for the simulator 3 2 3 Automatically Start Controllers The world concept should be changed from being just a json file to being a folder with the following items e JSON file describing the environment and objects within it e Controller folder containing the controller executables attached to this world These can be referenced by
5. asier to understand the structure of the code and start extending it to support features needed for new experiments The simulator has a fast forward function that will allow the user to speed up 2x and 3x real time It is important to note that controllers run asynchronously from the simulator and will get less computation time between frames to work 3 1 Benchmark The benchmark was run on a laptop with a dual core Intel i7 3520M processor running at 2 90GHz I let it run for about 5000 frames then discarded a couple of hundred frames at the start and end to avoid measuring unstable frames as it starts up and shuts down The goal is to keep each frame within 16 milliseconds allowing the simulation to run at real time speeds The world contained 200 robots with a controller each two small static obstacles in the middle and the walls surrounding the arena Table 3 1 contains the results of this benchmark 11 12 CHAPTER 3 RESULTS Figure 3 1 200 robot experiment Average 7 81 ms Stdev 0 94 ms Minimum 6 ms Maximum 35 ms Table 3 1 Time spent per frame with 200 robots and 200 controllers Data was collected over 3200 frames picked in the middle of a run to avoid warm up period 3 2 Future Work 3 2 1 Communication Between Robots The IR sensor class used in the Chirp class has a function objectHit that allows the user to get a pointer to the closest object a distance sensor can sense
6. ator I wanted to make a simulator capable of running hundreds of robots concurrently while being relatively simple to use and learn It needs to be accurate enough to be useful in experiments but simple enough to be able to handle many objects moving at once 1 2 Existing Simulators 1 2 1 Webots Webots was originally started as an attempt to improve the Khepera Simulator by adding realistic vision sensors and allowing for more robotics architectures 1 It comes with a few example worlds showing several robots in use like a gantry robot 2 CHAPTER 1 INTRODUCTION playing the towers of Hanoi a hexapod robot walking a swimming salamander robot a stewart platform robot and more 2 It supports writing controller code for both simulator and a physical robot through a single API compiling it to either target through makefiles linking against the correct library Webots comes with support for cross compiling code to multi ple existing robots like the e puck Khepera Hemisson LEGO Mindstorms and Aibo Users are able to extend this to their own robots providing an implemen tation of the API that runs on their design and a makefile able to cross compile to the architecture 3 Webots has a Fast2D simulation mode that uses the Enki simulator as a backend This removes a lot of the features Webots have but speeds up the simulation drastically by simplifying the environment being simulated Enki is sadly not integrated too well being
7. ction Model 13 3 2 5 More Sensor Types 0 0 00 0000004 14 3 2 6 Port Chirp Library to Simulator 14 Bibliography 17 Appendices 19 3 3 Documentation gt oos ce ssa oo se dass 19 3 4 Building and Running ChirpSim 19 341 Building ico ee e ee 19 3 42 Running za a bd dt ee 20 3 5 Controller Server API ma pea samea Emmen 20 3 6 Chirp Robot Controller API 2 2 n een 20 3 6 1 setspeed Message Details 20 3 6 2 getsensors Message Details 21 3 6 3 Example Controller in Python 21 dl World Format i A A OL 22 344 1 R ot Object io ie ra PE EG eS 22 Dro Chirp Object rra Lane a 22 3 08 Wall Objective Li 2 2220 a e Ge 23 3 7 4 PushableObject Object o o 24 Chapter 1 Introduction 1 1 Motivation The Chirp robot needed a simulator able to run a full swarm experiment without slowing down too much After looking at a few simulators capable of simulating custom robots I found a few issues stopping me from creating a model for the Chirp robot for these simulators Some simulators are simply too slow trying to be too realistic to be able to run many hundreds of robots concurrently in near real time Others are too complex to quickly jump into requiring the user to understand yet another layer of abstractions or requiring a complex setup to build and run the simul
8. eas only detectable by the sensor assigned to that type of area A line following sensor is slightly different in that the detectable area isn t a circle but instead a line segment The first type is the line segments on the ground to follow categorized so they only collide with line sensor bodies The other type of sensor are the line sensors themselves implemented through Box2D sensors to only detect collisions with the line segments See https www iforce2d net b2dtut collision filtering https www iforce2d net b2dtut sensors https www iforce2d net b2dtut raycasting and https www iforce2d net b2dtut world querying in particular to read more about sensors collision filtering and raycasting the world works in Box2D World querying is a more efficient way to do a raycast and is the preferred way to do so 3 2 6 Port Chirp Library to Simulator The Chirp robot will have a library for ease of programming it giving a much higher level view and access of the functionality in the robot The interface used by this library should be ported to a simulator library allowing users to use the exact same code for the simulator and the physical robot only needing to change the library linked to when compiling For a C implementation this would be relatively easy to do as the original library will be written in C The simulator version of the library would use the same headers but with a different implementation of all functio
9. ect Hosting The Meta Object System Documentation Qt Project 2013 URL http qt project org doc qt 4 8 metaobjects html visited on 06 16 2013 7 Julian Smart et al wxWidgets 2 8 12 A portable C and Python GUI toolkit 2011 URL http docs wxwidgets org stable visited on 06 16 2013 8 GNOME project Features gtkmm C Interfaces for GTK and GNOME 2013 URL http docs wxwidgets org stable visited on 06 16 2013 9 GNOME project libsigc The Typesafe Callback Framework for C 2012 URL https developer gnome org glibmm 2 34 visited on 06 16 2013 17 18 10 11 12 13 14 BIBLIOGRAPHY Martin Schulze and Murray Cumming libsigc The Typesafe Callback Framework for C 2013 URL https developer gnome org glibmm 2 34 visited on 06 16 2013 iMatix Corporation The Intelligent Transport Layer zeromg 2013 URL http www zeromq org visited on 06 16 2013 Erin Catto Box2D v2 2 0 User Manual 2013 URL http www box2d org manual html visited on 06 16 2013 Howling Moon Software Chipmunk2D Physics 2013 URL http chipmunk physics net visited on 06 16 2013 Howling Moon Software Chipmunk2D Physics Pro 2013 URL http chipmunk physics net chipmunkPro php visited on 06 16 2013 Appendices 3 3 Documentation The documentation has been attached along with this report placed in the docs folder of the zip
10. file This contains call caller graphs inheritance graphs include graphs and class diagrams It has been generated with Doxygen a standard utility for generating documentation of code 3 4 Building and Running ChirpSim 3 4 1 Building ChirpSim has the following requirements to build Box2D compiled for your platform as a static library then place the output files in the Libraries lib folder For windows the output files are Box2D lib and Box2Dd lib For OS X and Linux the output file is Box2D a Qt 5 for your platform Easiest way to get this set up is to download the in staller for your platform from http qt project org downloads This contains Qt Creator an IDE for C able to build and run Qt projects Open ChirpSim pro in Qt Creator and build in release mode The build button is on the bottom left corner of the IDE window 19 20 APPENDICES 3 4 2 Running After starting ChirpSim click the FileFile menu and select Load World In the ChirpSim win32bin examples folder there are three worlds containing roughly the same things but with differing amounts of robots Select one of these in the Open File dialog and put the window back in focus Yes it is a bug Now the world is loaded and you can move the camera using either the scrollbars or the arrow keys on the keyboard The Zoom menu lets you zoom in or out but scrolling the mouse wheel or using the trackpad scrolling gesture will also zoom There is also a file named
11. ildren lt objecti gt lt object2 gt lt object3 gt y Attributes children contains all the objects currently in the world 3 7 2 Chirp Object The Chirp object contains all the attributes required to create a Chirp object in the simulator It can be placed at different places in the world by setting the position attribute of the object Listing 3 2 Chirp object type Chirp a position ale TIO y 84 0 tigis Oy 3 7 WORLD FORMAT 23 angle 120 noisemax 0 1 Attributes string type must be Chirp not case sensitive object position has two attributes float x x component of the position float y y component of the position int id unique id of the robot used for sending messages to control the robot float angle starting angle of the robot float noisemax affects how much noise is applied to the wheel speeds during simulation 3 7 3 Wall Object The Wall object is a static body in the world impossible to move by any means It is created as a polygon body through vertices stored in the object These vertices must form a convex shape and be in counterclockwise order for the simulator to accept them To simulate a concave shape split it into multiple convex parts and put those in the hulls attribute Listing 3 3 Wall object al eos Wena e position seit gs SiO ae eal 0 Pe angle 0 Diyala Ds F ft els 10 Me 10 Ips if 2 10 yz 10 Iry T ele GO
12. k 3 6 2 getsensors Message Details Template lt id gt getsensors Parameters lt id gt id of the robot being controlled Returns ok lt sensorl gt lt sensor2 gt lt sensor8 gt where sensorl to sensor8 are the sensor values read from each sensor separated by a space 3 6 3 Example Controller in Python This controller simply sets the speed of the robot with id to random numbers between min_speed and maz_speed import random socket time def runbot id min_speed 10 max_speed 30 MOS S Mila 050 A port 37875 s socket socket socket AF INET socket SOCK_DGRAM speed_msg 0 setspeed 1 2 sensor_msg 0 getsensors while True left random randint min speed max speed right random randint min speed max speed s sendto speed msg format id left right host port tmp s recv 1024 s sendto sensor msg format id host port tmp s recv 1024 22 APPENDICES time sleep 0 016 sleep 16ms 3 7 World Format 3 7 1 Root Object The root object only contains the attribute children which is an array of objects in the world This allows for attributes to be set on the world at the root object while keeping it easy to access the objects under the world If the root object was the children array it would be necessary to parse whether each element in the array is an attribute or a physical object of the world Listing 3 1 Root object ch
13. le of running hundreds of robots concurrently at near real time each controlled by a separate controller I would like to thank my teaching supervisor Pauline Haddow Professor for her support and guidance throughout the project Robert Versvik Trondheim June 18 2013 iv Contents 1 Introduction VA Motivation et 22 halen Seah e rd ae 1 2 Existing Simulators e ata se aa 0 002 eee eee EXI Web ts 2 4 thee oe a 2 8 a oe ana eee Dh EO G 122 2 Enki a on as geen E A NR 1 23 Pl yer Stag 2 2 0 A near 1 3 The ChirpSim Project m a a tee ee en LIL WG alse 4 ti a AEG SE Be bt RR 2 Choices 2 1 Programming Language nn nn 21 1 SOARS TE DE SSE EG 212 Alternativ s Sein aa en HEG SN 2 2 Graphical User Interface Framework D DT Ouse ar Ara a Be GA EEE 22 2 Alternatives ar sek gates e Sa EO sale 3 2 3 Controller Architecture 23 1 UDP Socket 4 2 parese eee GE Ge Ge te se 2 3 2 Alternatives skadd ve rake e bee EN 2 4 Physical Simulation erida sekk esd Sarp alet Gas Pk ne ZA BOX2D 43 2 amp ho Dr Gi poe beh AE die Be vk RR Bs hin 2 4 2 Alternatives ee ee 2 5 Plugin Architecture ee ee ee 3 Results Sul Benchmark ses gerie e a AA AA a o dte 3 2 Future Work ch a foe lev a een ae ee 3 2 1 Communication Between Robots 3 2 2 World Editor vi CONTENTS 3 2 3 Automatically Start Controllers 13 3 2 4 Improve Fri
14. lemented as a plugin loaded dynamically from a plugin folder Qt has support for loading plugins at runtime through the class QPluginLoader as long as it implements an interface class available in the core This enables an application to search for plugins in a folder and load them all dynamically at run time For ChirpSim this would mean loading new physical objects through a plugin folder extending the simulator without having to recompile it all I decided against using plugins because any plugin has to be compiled once per platform it will be used on and against the specific version of Qt the core appli 10 CHAPTER 2 CHOICES cation uses This will become somewhat of a maintenance nightmare over time as plugins has to be updated to keep compatibility with the core application on new releases It s better to simply keep it as one application that is open sourced and open for contributions Chapter 3 Results The simulator is able to run many robots concurrently without slowing down much Figure 3 and table 3 1 show the performance of the simulator while running 200 robots simultaneously in a close quarters world where the robots collide with the walls or each other frequently The controller architecture ensures that many controllers can run asynchronously of the simulator only affecting the simulator through bottlenecking the CPU I have tried to keep the code relatively clean simple straight forward and docu mented so it is e
15. m containing a con troller for each robot that runs separate from the rest of the swarm For a simulator this means each robot must have at least one controller running with no method of directly reading any state but its own from the simulator Due to the amount of robots often employed in a swarm experiment it is unfeasible to run the controllers synchronously and sequentially with the simulation as this would scale poorly 2 3 1 UDP Socket The simulator will start a new thread listening on messages from a UDP socket forwarding any message read to the controller interface of a robot based on id The controller processes the message further before calling the correct setter or 8 CHAPTER 2 CHOICES getter to finish handling the message returning the string ok plus any data expected This is kept simple intentionally because I wanted the least amount of overhead for controllers as they communicate with the simulator I elected to use a UDP socket to read short messages over localhost on a separate thread from the simulation It is a very simple protocol that most languages have some support for which makes it possible to start creating controllers for the simulator relatively quickly 2 3 2 Alternatives Zero Message Queue is a library for implementing a fast message queue between threads processes and machines It is available in many languages works on most operating systems and implements multiple message carriers It tries t
16. many features like networking threading database connectors multimedia and camera playback unit testing along with classes to build a graphical user interface It also contains an implementation of JavaScript extended with support for signals and slots This allows applications to be scriptable having scripts react to events happening in the application 5 It extends C through a preprocessor they call the Meta Object System enabling signals and slots reflection gobject_cast which is a more efficient version of the standard C dynamic_cast 6 I chose Qt because of the QGraphicsView class and related classes a scene graph with built in scaling moving and rotation of the view Every node in the scene can respond to mouse events keyboard events when selected and more making it easier to interact with the scene compared to having to implement it all manually I can also create my own items for the scene that draw themselves to achieve the look I want 2 2 2 Alternatives wxWidgets is a cross platform library for building GUIs It is available for Windows OS X Linux and UNIX along with some mobile and embedded oper ating systems like iOS and Windows CE Along with the core library there are several libraries bundled with the download These include networking extra widgets for GUIs animations and widget layout making it a good alternative to Qt 7 2 3 CONTROLLER ARCHITECTURE 7 I chose Qt over wxWidgets because I was unable
17. ns in the cpp files This way when a user controller wants to compile against the simulator it 3 2 FUTURE WORK 15 can simply link against the simulator library Linking against the physical robot library will make the exact same code work on the robot instead For other languages the Simplified Wrapper and Interface Generator SWIG can be used to generate an interface those languages can use SWIG sup ports 19 languages as of June 18 2013 but an up to date list can be found at http www swig org compare html 16 CHAPTER 3 RESULTS Bibliography 1 Olivier Michel Webots Symbiosis between virtual and real mobile robots In Virtual Worlds Springer 1998 pp 254 263 2 Cyberbotics S r l Webots robot simulator Documentation User guide Section 3 1 2013 URL http www cyberbotics com guide section3 1 php visited on 06 16 2013 3 Cyberbotics S r l Webots robot simulator Documentation User guide Section 5 8 2013 URL http www cyberbotics com dvd common doc webots guide section5 8 html visited on 06 16 2013 4 Brian Gerkey Richard T Vaughan and Andrew Howard The player stage project Tools for multi robot and distributed sensor systems In Proceed ings of the 11th international conference on advanced robotics Vol 1 2003 pp 317 323 5 Qt Project Hosting Documentation Qt Project 2013 URL http qt project org doc visited on 06 16 2013 6 Qt Proj
18. o abstract away where the recipient of the message is whether it is over a local socket to another thread or process or to a completely different machine on the local network 11 I chose not to use this because it feels like overkill for the project especially when the simulator is intended to run on one computer with very simple interfaces for the controllers It also requires more setup compared to the UDP socket solution making it harder to jump in and start making controllers for the simulator See 11 for more information 2 4 Physical Simulation I decided to use a library for the physics of the simulation Some of the key features I was looking for were speed ease of use and amount of code examples and tutorials available The speed requirement enabled me to quickly discard physics engines made for 3D simulations as the extra dimension would go to waste and would increase the simulation complexity 2 4 1 Box2D Box2D is a physics engine written in C with a focus on speed It simulates a 2D environment with or without gravity filled with bodies and sensors that interact with each other 12 It has been used to simulate 2D environments in many applications requiring high speed per step making it a good match for 2 5 PLUGIN ARCHITECTURE 9 my goals with the simulator It has been ported to many languages e g Java C Python Javascript which means the simulator can be ported to another language without having to rewrite
19. oat density density of the object used to calculate mass float angle start rotation in degrees list hulls list containing hull objects which are lists of position objects object element position object containing an x and y coordinate float x x component of the position float y y component of the position
20. runbots py which implements a random controller setting random speeds and reading the sensor values continuously It accepts the number of robots to run either as a parameter when started or by manually typing it in when prompted To run it execute it as a normal python script 3 5 Controller Server API The controller server in ChirpSim listens to port 37875 only It will receive a message and read the ID number at the start then forward the rest of the message to the controller attached to the robot with the correct ID This will be the entry point for all controllers to communicate with any robot in the simulation 3 6 Chirp Robot Controller API The Chirp robot implements a very simple API with only two messages recognized at this time setspeed and getsensors The returned message will start with ok if there were no errors completing the request or error if an error happened The controller server on the simulator requires each message to be prepended with the id of the robot receiving the message This makes the connection completely stateless and helps make it easier to understand 3 6 1 setspeed Message Details Template lt id gt setspeed lt leftWheelSpeed gt lt rightWheelSpeed gt Parameters lt id gt id of the robot being controlled lt leftWheelSpeed gt speed desired on the left wheel 3 6 CHIRP ROBOT CONTROLLER API 21 lt right WheelSpeed gt speed desired on the right wheel Returns o
21. sed for real time games simulations and in software crunching numbers It is able to use many C libraries directly and can also export a C interface if needed for other languages to make use of There are also many C libraries and frameworks to help implement desired functionality like networking GUI and physics simulation It compiles into native code for the CPU which gives it high performance but forces it to be recompiled for each platform it targets ChirpSim has around 16 ms between each frame to simulate and draw the en vironment while handling messages sent by controllers to the robots C has libraries to solve that efficiently and quickly making it a good choice to achieve the requirements set Because of this I settled on C as my language of choice 5 6 CHAPTER 2 CHOICES 2 1 2 Alternatives Java is another language that can reach the speed requirements set by the program It runs on a virtual machine enabling it to run on multiple platforms without having to be compiled separately for each Java is slower than C but comes with a much larger standard library to help ease development I chose C over Java because the Qt framework gave me the convenience of the Java standard library along with QGraphicsView and its related classes allowing me to build a moving scene more easily than in Java 2 2 Graphical User Interface Framework 2 2 1 Qt Qt is a cross platform application and UI framework for C It contains
22. the JSON file and the simulator should automatically start the controller when the world is loaded Right now the world file has no way to say which controller is attached to each robot and assumes instead that the user starts the controllers manually If the simulator could automatically start the controllers it would be a big improvement to usability 3 2 4 Improve Friction Model Due to time constraints I was unable to model the friction in PushableObject correctly I set a constant 0 3f as friction on each part of the body which is used for both static and kinetic friction A proper simulation of the friction would model both static and kinetic friction and apply the correct force to the body to counteract any movement 14 CHAPTER 3 RESULTS 3 2 5 More Sensor Types The prototype only supports distance sensors This is enough to support the full functionality of the base Chirp model but there will be demand for other types of sensors like line following light heat or alcohol to mention a few Light heat and alcohol sensors can be implemented using very small circular Box2D sensor bodies to simulate the actual sensors then a bigger circular Box2D sensor body to simulate the area where detection is possible Calculating the distance from the sensor to the center of the detection area will give a scale that can be used to calculate the sensor value e g distance for light sensors Collision filtering will then be used to make the ar
23. the physics layer Box2D has an active community behind it helping new users through well written tutorials and helpful discussions on the official forums of Box2D This makes it very easy to get started on the physics engine and understand how to implement desired behavior See http box2d org for more information I chose Box2D because of the amount of example code and tutorials explaining the concepts available It is sufficiently fast has a mature API and is relatively easy to customize I also have more experience using it from past projects 2 4 2 Alternatives Chipmunk2D is an alternative 2D physics engine often used in real time en vironments It is based on an early version of Box2D but has since departed in many areas to distinguish it from Box2D 13 It is written in the C language enabling many languages to use it with relative ease without having to write a port It comes in a commercial pro version optimized for use on ARM processors and contains tools for generating collision geometry from images and procedural data 14 Neither of these is important for ChirpSim due to it running on x86 processors and only using relatively simple shapes for most objects I decided against Chipmunk2D because I found less tutorials easily available I prefer an object oriented API and I have less experience using it 2 5 Plugin Architecture I considered a plugin architecture for the root objects in the world so that each object is imp
24. to find a widget similar to the QGraphicsView of Qt This would mean I would ve needed to implement scaling rotation and translation of the scene myself along with mouse and keyboard handling for each item GTKMM isthe C bindings to the Gtk toolkit used in the GNOME desk top environment It only offers classes for making a GUI meaning the user has to find other libraries for any other functionality required by the application The GNOME project provides two companion projects in glibmm and libsigc 8 glibmm is a C wrapper around glib and provides threading networking smart pointers a main event loop asynchrous IO and more 9 libsige provides a sig nal slot system similar to Qt but implemented without a preprocessor step 10 A GUI can be quickly created in GLADE a software specialized in creating a GTK GUI and saving it as an XML file These XML files can be read by GTKMM to rebuild the GUI as designed in GLADE This makes it easier to rapidly evolve the user interface and allows stakeholders to see a close mockup of the user interface without having to flesh out the whole application It is also possible to create the GUI directly in code where needed I chose Qt over GTKMM and glibmm because it like wxWidgets lacks an answer to QGraphicsView It also fails to create a native looking GUI instead opting for its own look and feel 2 3 Controller Architecture A swarm of robots is a highly distributed computing platfor
25. trollers without being limited by the simulator itself It exposes all sensors and actuators through Player interfaces so that clients are able to run through these interfaces on both the simulator and the physical robot See 4 for more details on the Player Stage project 1 3 The ChirpSim Project ChirpSim is a simulator written for the Chirp robot developed for use in swarm experiments and aims to be capable of running several hundreds of robots simul taneously in a simple 2D environment 1 3 1 Goals The goals of the ChirpSim project are e The ability to run hundreds of Chirp robots concurrently in real time e A simulation that is quick but realistic enough to be useful to simulate experiments e A simple interface to code controllers against The questions I needed to answer were e Which programming language should I use e What can I use to create the user interface e How can I scale the controller system to be able to run hundreds of con trollers concurrently against robots in the simulation What are my options for simulating a physical world How can I enable extensibility CHAPTER 1 INTRODUCTION Chapter 2 Choices 2 1 Programming Language I had a couple of requirements to help me select a programming language for the simulator I wanted a language that can be used for writing high performance applications stable mature and that I have some experience in 2 1 1 C C is a language often u
26. unable to simulate many objects used in normal Webots because of how they are defined While the Fast2D mode definitively enables Webots to be used with swarm ex periments it is relatively hard to use The world can only have a restricted set of items in it for the simulator to recognize them otherwise they will be ignored Webots is also very expensive to license and with Fast2D it is merely a frontend for the free Enki simulator 1 2 2 Enki Enki is a fast 2D robot simulator created by Stephane Magnenat It is used in Webots as a Fast2D plugin but can be run alone as a standalone simulator Enki supports several robots out of the box like the Alice the e puck the khepera and the s bot It is also possible to extend Enki to add support for custom robots and the sensors used by this 1 2 3 Player Stage The Player Stage project consists mainly of two applications The Player robot device server and the Stage multiple robot simulator Player works as an abstraction layer over the hardware giving a predefined set of interfaces to interact with devices on a robot It contains client libraries in multiple languages to help write controllers for the robots but any language able to do TCP networking can communicate with Player 1 3 THE CHIRPSIM PROJECT 3 Stage is a multi robot simulator able to simulate a simple 2D environment The environment is simplified to be able to run many robots concurrently or run a few computation heavy con

View/Open

Contents

Download Pdf Manuals

Related Search

Related Contents