Multi-agent reinforcement learning: an approach based on agents' cooperation for a common goal

This paper is devoted to the problem of reinforcement learning in multi-agent systems. Multi-agent systems form a particular type of distributed artificial intelligence system. This paper presents an approach based on agents' cooperation for a common goal. By using other agents' experiences and knowledge, an agent may learn faster, make fewer mistakes, and create some rules for unseen situations. But the information communion among agents is deficient and limited. In this paper, we assume that every agent can only observe its neighbors' current positions and can see whether or not they reach the goal after the actions have been taken. Experimental results show the effectiveness of the proposed approach