Abstract
To flexibly coordinate a multi-agent system (MAS) composed of a large number of autonomous agents, we must allot the proper actions to the proper number of agents allocated to the proper areas. We have proposed a trial and error method to design a coordinating system composed of many agents communicating indirectly with diffusing signals, by modifying an agents' behavior rule. In this paper, we improved a simple sorting task model (ant-like robot, ALR) by using this method. We introduced an attracted move toward the signal source and a random move in desensitization into the agent's behavior rule of ALR. We named this signal-diffusion system. The signal-diffusion system operates according to the reaction-diffusion mechanism, and shows higher performance in the sorting task than ALR.
