Evolutionary Simulation for Co-Operative Behavior Learning on Multi-Agent Robots

Abstract

The purpose of our research is to confirm some behaviors of emergence appeared in distributed agents by simulating multi-agent robot systems in which each agent has an evolutionary algorithm. In this research, each agent robot has two behavior aims both collision avoidance from the other agent and target (food point) reaching motion for recovering self-energy. Evolutionary simulations add cooperative behavior learning of generating an environmental map for the above-mentioned multi-agent robots are discussed in this paper. Each agent robot has two conflicted tasks, that is, local individual behaviors for the self-protection and global group behaviors for the cooperative task. Furthermore, it has the additional algorithm of the group evolution that parameters of the best agent are copied to a dead agent which lost its energy. We also performed some simulations to find how much the rate of individual tasks and cooperative tasks influence on the group behaviors by using our evolutionary behavior learning simulator for generating the environmental map. As a result, we confirmed that the balanced weight between cooperative and individual parameters produces excellent cooperative behaviors in the homogeneous agent group.