1. Meeting report NTU CSIE Adviser ： Prof. Jane Hsu Speaker ： Wen-Chieh Fang 2005/09/21

2. Agenda  Simulation Scenario  Behavior-based Reactive Paradigm  Problem description  Robot model  Random walk  Coordination  Area bounding  Emergency detecting  Demo  Summary and future work

3. Simulation scenario  Surveillance task ： Three robots on their area to survey emergency （ red color object ）  Robots are behavior- based （ random-walk, coordination, area bounding,emergency detecting ）

4. Behavior-based Reactive Paradigm  Vertical decomposition of tasks PLANSENSEACT

5. Problem description  Project ： ITRI plan  Problem ： Multi-robot coordination  Assumption ：  Robots are homogeneous  Robots can communicate without lost information  Robots have no model/knowledge of other robots  Robots have no ability of self-localization

6. Robot model  Koala robot ：： Highly modular, all terrain indoor robot  Sensors ： 16 Infra- red proximity and ambient light sensors  Accessories ： a camera, an Infra-red ground proximity sensor Adopted from: http://www.cyberbotics.com

7. Random walk  Algorithm ： Braitenberg vehicle approach  Feature ：  Relying on the chances to find targets  Different from complete approach to canonical clean-floor task （ mainly for stationary targets ） Adopted from [Acar et. al. 2003] Random walk Back & forth

8. Coordination  Messaging  “inhibition” message  State  none → “commitment” state  Assumption  At most two robots tackle the same emergency

9. Area bounding  Each robot is equipped with an distance sensor below the base of robot.  There are landmarks on the bounding of surveillance area

10. Emergency detecting  Device ： Camera  Assumption ： we use stationary red color objects in place of emergencies such as intruders or fire.

11. Demo

12. Summary & future work  Make devices such as camera, emitter and receiver accord with the physical models in real world (for example: device size, measurement constraints or limits).  In the random walk algorithm, extend the behaviors that maximize the distance between adjacent robots so that the robots have more chances to find the emergencies.  Fault tolerance or uncertainty problems such as communication lost, sensing noise etc.  Search multiple moving targets.  Mission support ： remainders can change their surveillance area to compensate the lacking of committing robots.

13. Reference  [Acar et. al. 2003] Ercan U. Acar, Howie Choset, Yangang Zhang, and Mark Schervish, “Path Planning for Robotic Demining: Robust Sensor-based Coverage of Unstructured Environments and Probabilistic Methods “, The International Jounal of Robotics Research, Vol. 22, No. 7-8, July-August 2003, pp. 441-466. Ercan U. Acar, Howie Choset, Yangang Zhang, and Mark Schervish, “Path Planning for Robotic Demining: Robust Sensor-based Coverage of Unstructured Environments and Probabilistic Methods “, The International Jounal of Robotics Research, Vol. 22, No. 7-8, July-August 2003, pp. 441-466.