COLLECTIVE ROBOTICS FOR RESOURCE DISTRIBUTION

1
COLLECTIVE ROBOTICS FORRESOURCE DISTRIBUTION

• Steven de Jong
• steven.dejong_at_cs.unimaas.nl
• MICC Ph.D. Day
• 15 december 2005

2
Storage facilities
3
Resource distribution

• EASY?
• Bring objects from A to B
• COMPLEX? Weyns
• Objects can have priorities
• Object offerings are often irregular
• Environment can change
• Facilities are very large
• REQUIREMENTS
• Adaptive, functional, robust scalable

4
Classical approaches

• Classical approaches
• Multi-Agent Planning De Weerdt
• Contract Net FIPA
• Problems Dorigo, Wolpert, Stone,
• Have central control
• and/or
• Rely heavily on inter-agent messaging
• Complexity explosion
• Number of robots
• Number of tasks

?
5
Collective systems in biology Dorigo
6
How do they do it? How do we?

• Ant hills ? mechanisms ? storage facilities
• Key mechanisms
• Communication through environment
• Pheromones, et cetera
• Collective/externalized memory (configuration)

Please dont use electric cattle when microwave
is turned on.
7
Our research track (1)

• Central agent
• Grid world
• Potential field
• Reactive robots

8
Our research track (2)

• Knowledge and memory distributed from central
  agent
• To individual robots
• To the environment (communication)
• Key question
• How can a group of co-operative agents obtain
  co-ordination in a robust way, in an
  ever-changing environment?
• Approach
• Collective intelligence, RL, sensor networks,

9
Summary

• Problem
• Systems with many robots either emergent or
  predefined
• Wish to have adaptive and reliable large
  systems that perform difficult tasks predictably
• Solution
• Methodology required function of the group ?
  requirements for individuals and environment
• Straightforward (e.g., reactive) specification
  for individuals

10
References (selection)

• Danny Weyns et al Gradient field-based task
  assignment in an AGV transportation system
• FIPA www.fipa.org
• Marco Dorigo / IRIDIA iridia.ulb.ac.be/mdorigo/
  ACO/ACO.html

• David Wolpert et al General Principles Of
  Learning-Based Multi-Agent Systems
• Peter Stone et alMulti Robot Learning for
  Continuous Area Sweeping
• Mathijs de Weerdt Multi-agent Planning An
  introduction to planning and coordination.