Control of Autonomous Swarms Using Gibbs Sampling

A distributed control approach is proposed for self-organization of autonomous swarms. The swarm is modeled as a Markov random field (MRF) on a graph where the (mobile) nodes and their communication/sensing links constitute the vertices and the edges of the graph, respectively. The movement of nodes is governed by the Gibbs sampler. The Gibbs potentials, local in nature, are designed to reflect collective goals such as gathering, dispersion, and linear formation. The algorithm can be run completely in parallel, and hence it is robust and scalable. Simulation results are provided to illustrate the proposed method.