Distributed Particle Swarm Optimization - particle allocation and neighborhood topologies for the learning of cooperative robotic behaviors

摘要: In this article we address the automatic synthesis of controllers for coordinated movement multiple mobile robots, as a canonical example cooperative robotic behavior. We use five distributed noise-resistant variations Particle Swarm Optimization (PSO) to learn in simulation set 50 weights an artificial neural network. They differ on way particles are allocated and evaluated how PSO neighborhood is implemented. addition, centralized approach that allows benchmarking with versions. Regardless learning approach, each robot measures locally individually performance group using exclusively on-board resources. Results show four obtain similar fitnesses version, always able learn. The other variation fails properly some runs, results lower fitness when it succeeds. test systematically learned real experiments.