The TeamBots Environment Tucker Balch The Borg Lab Georgia Institute of Technology Wonderful to be here! This is work led by me in collaboration with my graduate students at CMU I am going to focus on work begun over the last year that I plan to continue in the future General -- collaborations
Why TeamBots? Why “Environment?” Robotics researchers need more than a language, we need a flexible, manageable environment that provides: Consistent APIs to robot hardware Simulation Communication Graphical tools Code reuse Tucker Balch Georgia Institute of Technology Mobile Robot Programming 10 May 2002
What is TeamBots? TeamBots is a Java-based collection of applications and libraries designed to support robotics research: TBSim: configurable simulation tool TBHard: robot executive RoboComm: communications package Clay: library for programming behavior-based controllers (maybe Clay is an architecture) Tucker Balch Georgia Institute of Technology Mobile Robot Programming 10 May 2002
TeamBots Software Architecture: Design Robot Controller API Simulation Hardware Tucker Balch Georgia Institute of Technology Mobile Robot Programming 10 May 2002
Java: The Good Syntax supports layered control system configuration (more on that later) Rich set of libraries (threads, GUI tools, communications) Portable Automated documentation Hard to shoot yourself in the foot Strongly typed Object oriented No pointers Tucker Balch Georgia Institute of Technology Mobile Robot Programming 10 May 2002
Java: The Bad Religion & Hype Java is slow Timing is unpredictable due to GC Work arounds: Speed: Use JITs, native compilers GC GC at regular intervals 10% to 20% performance hit er Tucker Balch Georgia Institute of Technology Mobile Robot Programming 10 May 2002
Example TeamBots Simulations RoboCup small size soccer Nomad 150 Probotics Cye vehicle Outdoor vehicles Robot Controller API Simulation Tucker Balch Georgia Institute of Technology Mobile Robot Programming 10 May 2002
How TBSim Works Read in and parse description file: Two types of objects Objects without control systems Objects with control systems (robots) For each object: object.init() For each control system: cs.init() While not done For each object: object.takeStep() For each control system: cs.takeStep() Tucker Balch Georgia Institute of Technology Mobile Robot Programming 10 May 2002
TBSim Implementation Control Systems Simulated World Robot 1 Control System Robot 2 Control System simulated obstacle Simulated World simulated obstacle simulated robot hardware simulated obstacle simulated robot hardware simulated obstacle Tucker Balch Georgia Institute of Technology Mobile Robot Programming 10 May 2002
Description File Syntax bounds –5 5 5 –5 // meters timestep 100 // milliseconds timeout 60000 // milliseconds trials 10 graphics on seed 993 windowsize 500 500 // pixels object Obstacle 0 4 0 1 x0000FF x000000 4 robot Nomad150 forage 0 1 0 x000000 xFF0000 2 Tucker Balch Georgia Institute of Technology Mobile Robot Programming 10 May 2002
TeamBots Hardware Support Nomad 150 (Balch & Arkin, Georgia Tech) ISR Pebbles (Ram, Georgia Tech) Probotics’ Cye (Balch & Veloso, CMU) Amigobot (Luke, UMD) RWI ATRV (Koenig & Balch, Georgia Tech) Robot Controller API Hardware Tucker Balch Georgia Institute of Technology Mobile Robot Programming 10 May 2002
TBHard java TBHard Nomad150 forage desoto.cc.gatech.edu 3 600 0 0 0 Tucker Balch Georgia Institute of Technology Mobile Robot Programming 10 May 2002
Nomad 150 Balch, AI Magazine, 1997. Balch, Autonomous Robots, 2000. Tucker Balch Georgia Institute of Technology Mobile Robot Programming 10 May 2002
Nomad 150 Balch, AI Magazine, 1997. Balch, Autonomous Robots, 2000. Tucker Balch Georgia Institute of Technology Mobile Robot Programming 10 May 2002
Nomad 150 Balch, AI Magazine, 1997. Balch, Autonomous Robots, 2000. Tucker Balch Georgia Institute of Technology Mobile Robot Programming 10 May 2002
Nomad 150 Balch, AI Magazine, 1997. Balch, Autonomous Robots, 2000. Tucker Balch Georgia Institute of Technology Mobile Robot Programming 10 May 2002
Probotics’ Cye Tucker Balch Georgia Institute of Technology Mobile Robot Programming 10 May 2002
How Robot APIs are Defined Make use of Java features Inheritance Interfaces Robot Controller API Simulation Hardware Tucker Balch Georgia Institute of Technology Mobile Robot Programming 10 May 2002
Robot API Hierarchy: Design Simple extends Nomad150 Nomad150Hard Nomad150Sim implements extends Nomad150Comm Nomad150CommSim Nomad150CommHard implements Tucker Balch Georgia Institute of Technology Mobile Robot Programming 10 May 2002
Robot API Hierarchy: Code public interface Nomad150Comm extends Nomad150, Transciever {} public class Nomad150CommSim extends Nomad150Sim implements Nomad150Comm, SimulatedObject Tucker Balch Georgia Institute of Technology Mobile Robot Programming 10 May 2002
Inter-Robot Communication: RoboComm Simple API to TCP/IP Unicast Broadcast Multicast Implemented in simulation and on mobile robots Uses Java serialization for marshaling and unmarshaling Tucker Balch Georgia Institute of Technology Mobile Robot Programming 10 May 2002
Inter-Robot Communication t.unicast(2, new stringMessage( "hello!")); if (r.hasMoreElements()) new_message = r.getNextElement(); Tucker Balch Georgia Institute of Technology Mobile Robot Programming 10 May 2002
Clay: An Architecture for Robot Control Uses features of Java syntax to embed perceptual processes within action and selection processes Allows specification of flexible hierarchies Run time execution is efficient because only the necessary portions of the configuration are executed Includes library of perceptual and motor schemas Tucker Balch Georgia Institute of Technology Mobile Robot Programming 10 May 2002
Clay: Execution Hierarchy Hardware/Simulation Tucker Balch Georgia Institute of Technology Mobile Robot Programming 10 May 2002
Building Blocks: Motor Schemas Multiple independent processes each generate a vector combined by weighted summation Computationally simple and fast Enables design by composition. (Arkin 1989) Related to artificial potential fields Khatib (85), Krogh (84), Payton (89), Singh (98) Tucker Balch Georgia Institute of Technology Mobile Robot Programming 10 May 2002
Motor Schemas: Move to Goal Tucker Balch Georgia Institute of Technology Mobile Robot Programming 10 May 2002
Motor Schemas: Avoid Obstacle Tucker Balch Georgia Institute of Technology Mobile Robot Programming 10 May 2002
Motor Schemas: Avoid Obstacle + Move to Goal Tucker Balch Georgia Institute of Technology Mobile Robot Programming 10 May 2002
Example: Behaviors for Pushing Tucker Balch Georgia Institute of Technology Mobile Robot Programming 10 May 2002
Specification at Initialization Time detect_home = new v_Goal_r(abstract_robot,0,0); move_to_home = new v_Attraction_v(detect_home); detect_obstacles = new va_Obstacles_r( abstract_robot); avoid_obstacles = new v_Avoid_va(2.0, 1.0, detect_obstacles); swirl_obstacles = new v_Swirl_va(2.0, 1.0, detect_obstacles, detect_goal); move_to_home swirl_obstacles avoid_obstacles detect_home detect_obstacles Tucker Balch Georgia Institute of Technology Mobile Robot Programming 10 May 2002
Methods of Composition Weighted sum Winner take all Perceptual sequencing Learning Tucker Balch Georgia Institute of Technology Mobile Robot Programming 10 May 2002
Example: Combining by Weighted Sum avoid_n_swirl = new StaticWeightedSum_va(); avoid_n_swirl.embedded[0] = avoid_obstacles; avoid_n_swirl.weights[0] = 0.5; avoid_n_swirl.embedded[1]= swirl_obstacles; avoid_n_swirl.weights[1] = 0.5; avoid_n_swirl.embedded[2] = move_to_home; avoid_n_swirl.weights[2] = 1.0; steering_configuration = avoid_n_swirl; avoid_n_swirl move_to_home swirl_obstacles avoid_obstacles detect_home detect_obstacles Tucker Balch Georgia Institute of Technology Mobile Robot Programming 10 May 2002
Communication as Sensing detect_ball = new v_DetectRed_r(abstract_robot); team_ball_obs = va_CommRed_r(abstract_robot); ball_observations = v_Combine_vav(team_ball_obs, detect_ball); fused_ball_obs = v_Fuse_va(ball_observations); fused_ball_observaions ball_observations detect_ball team_ball_obs Tucker Balch Georgia Institute of Technology Mobile Robot Programming 10 May 2002
Comments on Clay/Java Nodes (schemas) are naturally embedded, combined and selected using Java syntax Java provides type checking at configuration time For node configuration For robot/control system matching Tucker Balch Georgia Institute of Technology Mobile Robot Programming 10 May 2002
Research & Education Using TeamBots Robot formations (Balch & Hybinette) Learning behaviors for soccer & foraging (Balch) Cooperative observation and localization (Stroupe & Balch) Learning behaviors for herding (Potter, des Jardins) Pheromone-based behavior (Payton) Robot soccer (Balch, Kitano) Education: SoccerBots Tucker Balch Georgia Institute of Technology Mobile Robot Programming 10 May 2002
Obtaining TeamBots www.teambots.org Free for non-commercial use New release due June 1 Tucker Balch Georgia Institute of Technology Mobile Robot Programming 10 May 2002
Summary TeamBots architecture leverages OO/Java features to provide: Rapid prototyping in simulation Using well-defined API to robot hardware Behavior specification using Clay (or not) Easy to use robot-robot communication Tested control systems run directly on robots Tucker Balch Georgia Institute of Technology Mobile Robot Programming 10 May 2002