A reinforcement learning system for swarm behaviors

This paper proposes a neuro-fuzzy system with a reinforcement learning algorithm to realize speedy acquisition of optimal swarm behaviors. The proposed system is constructed with a part of input states classification by the fuzzy net and a part of optimal behavior learning network adopting the actor-critic method. The membership functions and fuzzy rules in the fuzzy net are adaptively formed online by the change of environment states observed in trials of agentpsilas behaviors. The weights of connections between the fuzzy net and the value functions of actor and critic are trained by temporal difference error (TD error). Computer simulations applied to a goal-directed navigation problem using multiple agents were performed Effectiveness of the proposed learning system was confirmed by the simulation results.