Abstract
In the multi-agent systems, plural autonomous agents are expected to be able to learn the cooperative behavior. In a learning environment with only simple cooperative task, based on the reward redistribution between agents, we have already proposed a construction method of the multi-agent systems with autonomous group creation ability, which is able to strengthen the cooperative behavior of the group. But in case of a learning environment with also the competition among agents, we can not adopt this method because interests of each agent might conflict. Therefore, in order to acquire the appropriate group behavior for all agents by balancing the behavior of competition and cooperation autonomously, it is necessary to improve the social interaction among agents. In the concurrent learning environment with both competitive and cooperative behavior, we proposed a construction of autonomous learning mechanism by using the reinforcement learning through the social interaction among agents.
