Achieving corporative behavior in heterogeneous agents using hierarchic reinforcement learning-an approach to piano mover´s problem

Our approach is to achieve the cooperative behavior of autonomous decentralized agents constructed with Q-Learning, which is a type of reinforcement learning. The piano mover´s problem is employed. We propose the multi agent architecture that has an external agent and internal agents. Internal agents are heterogeneous and they can communicate with each other. The movement of the external agent depends on the composition of the actions of internal agents. According to learning its own shape by internal agents, it is expected that the agents avoid obstacles. We simulate our method on a two-dimensional continuous world. The results show the effect of our method.