In this paper a novel, Gibbs sampler-based algorithm is proposed for coordination of autonomous swarms. The swarm is modeled as a Markov random field (MRF) on a graph with a time-varying neighborhood system determined by local interaction links. The Gibbs potential is designed to reflect global objectives and constraints. It is established that, with primarily local sensing/communications, the swarm configuration converges to the global minimizer(s) of the potential function. The impact of the Gibbs potential on the convergence speed is investigated. Finally a hybrid algorithm is developed to improve the efficiency of the stochastic scheme by integrating the Gibbs sampler-based method with the deterministic gradient-flow method. Simulation results are presented to illustrate the proposed approach and verify the analyses.