Collective Robotics. Marcin Pilat

Transcription

1 Collective Robotics Marcin Pilat

2 Introduction Painting a room Complex behaviors: Perceptions, deductions, motivations, choices Robotics: Past: single robot Future: multiple, simple robots working in teams

3 Collective Robotics Teams of robots working collectively to solve a problem Non-cooperative vs. cooperative Observations of social insects Pasteels et al. [From Individual to Collective Behavior in Social Insects, pages , 1987] collective behavior is not simply the sum of each participant s behavior, as others emerge at the society level

4 Topics Flocking Group Behavior Robot Formations Small Unit Operations ALLIANCE System Cooperative Transport

5 Flocking Boids Craig Reynolds, 1986 Basic Flocking Model Separation Alignment Cohesion

6 Flocking Boid reacts only to local neighborhood Complex models Obstacle avoidance Goal seeking Excellent example of emergence Chaotic + ordered = life-like Unpredictable over moderate time scales

7 Flocking Applications Animated short (SIGGRAPH 87) Stanley and Stella in: Breaking the Ice Movies 1992 Tim Burton film Batman Returns computer simulated bat swarms and penguin flocks

8 Group Behaviors Animation Lab Georgia Tech Herds, flocks, schools Control algorithm for motion of robot in a herd Dynamic simulation of physical robots in herds Result: natural looking motion for walking, running, climbing; obstacle avoidance, grouping, rough terrain locomotion

9 Group Behavior Sample results One legged robots hopping [front] [side] Bicycle race obstacle avoidance [front] [side] Bicycle race group turns [1] [2]

10 Robot Formations Interaction Lab: USC Robotics Labs Behavior-based systems (BBS) Flexile, robust, scalable multi-robot spatial formations using local sensing and control Formation goal: N mobile robots establishing and maintaining some predetermined geometric shape: column, diamond, wedge, line

11 Behavior-Based Control Consist of a collection of behaviors Behavior Properties Processes or control laws Achieve and/or maintain goals Properties of BBSs React in real-time Uniform structure and representation

12 Robot Formations Equipment: sonar, laser, camera, radio link Keep single friend at desired angle Only heart-beat communication Algorithms validated in simulation and physical robots Formation Video Basic Behavior Video [Dr. M. Mataric, USC]

13 Small Unit Operations USC Robotics Research Laboratory & DARPA Autonomous taskable vehicles Ground based (UGVs) Airborne (UAVs) Equipped with architecture enabling online-learning and reconfiguration Ability to respond to situational and environmental variables

14 Small Unit Operations Scenario Robot colony in realistic reconnaissance scenario Multiple UAVs identify and hover in formation over a ground target UGVs use obstacle avoidance and formation algorithms to navigate to the goal UGVs enter goal building and send back visual images of interior

15 Small Unit Operations Success More than one UGV able to enter the goal building and send back visual imagery from within Perimeter Protection Task 2 UGVs Pioneer model 1 UAV AVATAR model

16 ALLIANCE Cooperative Robotics: Oak Ridge National Lab. ALLIANCE software system Motivates robots to carry out group missions and make adaptations to work environment Robots select actions based on mission progress and actions of teammates Enables robots themselves to respond to failures

17 ALLIANCE Motivational Behaviors Impatience drives robot to complete a task that is not being completed by other robots Acquiescence allows robots to give up tasks they cannot complete successfully

18 ALLIANCE ALLIANCE Software System No central command Robots broadcast periodic messages indicating what they are doing Sensory feedback used to measure progress on each task

19 ALLIANCE Applications Hazardous waste cleanup [1] [2] [3] Surveillance and monitoring Military applications (location of land mines) ALLIANCE system results Movement in formation Box pushing [homogeneous] [heterogeneous] Cooperative observation of multiple targets Cooperative Baton passing

20 Cooperative Transport CRIP: University of Alberta [Kube] Swarm-based robotics Insect colony: decentralized, flexible, robust Drawbacks: stagnation, robot programming Cooperative Box Pushing Collective Prey Retrieval in ants

21 Cooperative Prey Retrieval Observed in many species of ants Solitary vs. Group Transport Increased efficiency Solitary to Group Transport Ant tries to carry item Ant tries realigning and repositioning Ant recruits nestmates

22 Cooperative Prey Retrieval Nestmate Recruitment: short or long range Short-range recruitment (SRR) Poison gland secretion release after discovery Detectable up to 2m away Long-range recruitment (LRR) Chemical trail of poison gland secretions to the nest Nestmates leave the nest and follow the trail

23 Cooperative Prey Retrieval Coordination Example of stigmergy Stagnation Recovery If no progress in transport task Ants use realignment and repositioning Sorting-out behavior before transport moves

24 Cooperative Box Pushing Kube, Zhang Objective: Locate a brightly lit box Move box to a goal location Robots: 3 sensors: goal, robot, obstacle 2 actuators: left, right wheel motors G r B r r r r

25 Cooperative Box Pushing Behaviors: FIND default; explore FOLLOW follow closest robot SLOW slow motor speed GOAL move toward goal AVOID move away from obstacle Subsumption or Adaptive Logic Network

26 Cooperative Box Pushing Stagnation Recovery Stagnation detection Realign small random change in pushing angle Reposition change point of contact with box Observed increase in success percentage with stagnation recovery behaviors

27 Cooperative Box Pushing Tests: First tested on a simulator Physical test on 5 robots (undirected) Directed test on physical robots Multiple goal direct test

28 Cooperative Box Pushing Results Simple cooperation without direct communication using stigmergy can accomplish the box pushing task Movie

29 Conclusion Collective Robotics a HOT & FUN area Current research on simple problems Future: micro-machines Important applications Microsurgery, waste disposal, warfare, exploration Increasingly important

30 References Flocking / Group Behaviors: Robot Formations / Small Unit Operations ALLIANCE: Cooperative Transport: Swarm Intelligence Bonabeau, Dorigo, Theraulaz

31 Other Videos Miscellaneous Videos: On the Run [MIT AI Lab] Trail Following [USC Robotics Lab]