A Shared Experience Tuples Multi-Agent Cooperative Reinforcement Learning Algorithm
Chang Wang · 2005
Q--learning algorithm is one of the most popular model--free reinforcement learning algorithms. This paper extends Q--learning algorithm to multi--agent cooperative team domain properly, and proposes a shared experience tuples multi--agent cooperative reinforcement learning algorithm, in which a new knowledge representation form composed of sequential pair as 〈state-- value, action--valu〉 is proposed. And experience tuples are shared with other agents through similarity transformation according to homogeneous subtasks. In the end, the algorithm is applied to pursuit game domain. The result shows the validity of the algorithm which can speed up the progress of pursuit task and is beneficial to the accomplishment of cooperative tasks and improvement of cooperative effectivity.