Improvement with Joint Rewards on Multi-agent Cooperative Reinforcement Learning
Pan Ying, Dehua Li · 2008
Cooperation among agents is important for multi-agent system. In this paper, an improved cooperative reinforcement learning algorithm is proposed, which based on joint rewards to insure agents to learn cooperative behavior. Furthermore, a symmetry idea is included in the algorithm to reduce the states size of reinforcement learning. The experiment results show the efficiency and well convergence of the algorithm.