Structured and Stochastic Rules for Motions of Groups of Robotic Agents

John Baillieul
Boston University
Aerospace & Mechanical Engineering

The talk will briefly describe recent research on decentralized control of large groups of mobile agents. Both highly structured
motions as well as "softly structured" motions in which each agent
moves in accordance with probabilistic rules will be considered. In
all cases, patterns of sensing and interagent communication are
naturally described in the language of graph and lattice theory.
The goal of the control techniques is the parsimonious use of both
sensor and communicated information. While highly structured motions
seem to involve challenging complexities of scale, local rules
inspired by statistical mechanics and quantum Heisenberg models
provide an approach to effective use of large groups of agents for
applications of search and surveillance. In this regard, a simple
abstraction in which a building interior is represented as a
connected graph is presented. Search and surveillance algorithms may
then be described as dynamical systems on these graphs. It is shown
how purely local, decentralized rules for random movement will lead
to a random walk in which there is a prescribed likelihood of
visitation to each part of the graph. It is shown how uniform
coverage can be achieved most rapidly and how the parallelism
afforded by multiple agents can be most effectively utilized.

Presentation Files (Zip Archive)

Back to Swarming by Nature and by Design