Multi-agent reinforcement learning: an approach based on the other agent´s internal model

In a multi-agent environment, whether one agent´s action is good or not depends on the other agents´ actions. In traditional reinforcement learning methods, which assume stationary environments, it is hard to take into account of the other agent´s actions which may change due to learning. In this article, we consider a two-agent cooperation problem, and propose a multi-agent reinforcement learning method based on estimation of the other agent´s actions. In our learning method, one agent estimates the other agent´s action based on the internal model of the other agent. The internal model is acquired by the observation of the other agent´s actions. Through experiments, we demonstrate that good cooperative behaviors are achieved by our learning method