抄録
It is thought that the cooperative action by humanoid robots will have an important meaning because they have flexible bodies which enable them to operate various motions. But self-conditions of the humanoid robots influence their actions, and it causes some troubles when they act, moreover when they cooperate. In this paper, we propose the reinforcement learning system based on integration of multi-humanoids senses. Constructing state space by using this sensory integration, robots learn the optimal movements. We expect that it makes them cover the weakest robot, solves the problems between the individual differences of robots, and enables them to achieve the tasks more flexibly. Using real humanoid robots, we certify the validity of the proposed technique.
