Emergence of Cooperative Behavior based on Learning and Evolution in Collective Autonomous Mobile Robots

Abstract

In this paper, we propose a behavior learning algorithm of the collective autonomous mobile robots based on the reinforcement learning and conditional evolution. The cooperative behavior is a high level phenomenon observed in the society of social animals and, recently the emergence of cooperative behavior in collective autonomous mobile robots becomes an interesting field in artificial life. In our system each robot with simple behavior strategies can adapt to its environment by means of the reinforcement learning. The internal reinforcement signal for the reinforcement learning is generated by fuzzy inference engine, and dynamic recurrent neural networks are used as an action generation module. We propose conditional evolution for the emergence of cooperative behavior. The evolutionary conditions are spatio-temporal limitations to the occurrence of genetic operations. We show the validity of the proposed learning and evolutionary algorithm through several computer simulations.