Path Planning for Multi Agent Systems

1
Path Planning forMulti Agent Systems

• by
• Kemal Kaplan

2
Multi Agent Systems (MAS)

• A multi-agent system is a system in which there
  are several agents in the same environment which
  co-operate at least part of the time.
• Complexity of the path planning systems for MAS
  (MASPP) increase exponentially with the number of
  moving agents.

3
Problems with MASPP

• Possible problems of applying ordinary PP methods
  to MAS are,
• Collisions,
• Deadlock situations, etc.
• Problems with MASPP are,
• Computational overhead,
• Information exchange,
• Communication overhead, etc.

4
Classification of Obstacles

• Usually other agents are modelled as unscheduled,
  non-negotiable, mobile obstacles in MASPPs.
• Category of Obstacles from Arai et. al. (89)

5
Proposed Techniques

• Centralised Approaches
• Decoupled Approaches
• Combined Techniques

6
Centralised Approaches

• All robots in one composite system.
• Find complete and optimum solution if
  exists.
• Use complete information
• - Computational complexity is exponential w.r.t
  the number of robots in the system
• - Single point of failure

7
Decoupled Approaches

• First generate paths for robots (independently),
  then handle interactions.
• Computation time is proportional to the
  number of neighbor robots.
• Robust
• - Not complete
• - Deadlocks may occur

8
Combined Techniques

• Use cumulative information for global path
  planning, use local information for local
  planning
• Think Global Act Local

9
Utilities For Combined Techniques

• Global Planning Utilities
• The aim is planning the complete path from
  current position to goal position.
• Any global path planner may be used. (e.g. A,
  Wavefront, Probabilistic Roadmaps, etc.)
• Requires graph representation achieved by cell
  decomposition or skeletonization techniques.

10
Utilities For Combined Techniques (II)

• Local Planning Utilities
• The aim is usally avoid obstacles. However,
  cooperation should be used also.
• Any reactive path planner can be used. (e.g.
  PFP, VFH, etc.)
• No global information or map representaion
  required. Decisions are fast and directly
  executable.

11
Improvements for Combined Techniques

• Priority assignment
• Aging (e.g. the forces in a PFP varies in case of
  deadlocks)
• Rule-Based methods (e.g. left agent first, or
  turn right first)
• Resource allocation (leads to suboptimal
  solutions)

12
Improvements for Combined Techniques (II)

• Robot Groups
• A leader and followers
• Many leaders (or hierarchy of leaders and
  experience)
• Virtual leader
• Virtual dampers and virtual springs
• Assigning dynamic information to edges and
  vertices

13
Possibe MAS environmets for MASPP

• Robocup 4-Legged League
• Robocup Rescue
• SIMUROSOT, MIROSOT (?)
• Games (RTS, FPS)
• ...

14
MASPP Example ARAI OTA 89

• Measures
• Computational Load
• Total length of the generated trajectories
• The radius of curvature of the generated
  trajectories
• Total motion time
• Preferred measure is the first one

15
MASPP Example ARAI OTA 89

• Properties of agents

16
MASPP Example ARAI OTA 89

• Problem 1

17
MASPP Example ARAI OTA 89

• Problem 2

18
MASPP Example ARAI OTA 89

• Virtual Impedance Method

19
MASPP Example ARAI OTA 89
20
MASPP Example ARAI OTA 89
21

• Questions?
• kaplanke_at_boun.edu.tr