共享经验的多主体强化学习研究

Research on Multi-agent Reinforcement Learning with Sharing Experience

合作多主体强化学习的关键问题在于如何提高强化学习的学习效率。在追捕问题的基础上,该文提出一种共享经验的多主体强化学习方法。通过建立合适的状态空间使猎人共享学习经验,根据追捕问题的对称性压缩状态空间。实验结果表明,共享状态空间能够加快多主体强化学习的过程,状态空间越小,Q学习算法收敛越快。

How to improve the efficiency of reinforcement learning is the key problem of reinforcement leaning with multi-agent collaboration. This paper proposes a method of multi-agent reinforcement learning with sharing experience based on the research to pursuit problem. By applying this method the hunters can share the learning experience through constructing the appropriate state space. It further compresses the state space according to the symmetry character of pursuit problem. Experimental results show that sharing state space can expedite the process of multi-agent reinforcement learning. The smaller the state space is, the faster Q learning algorithm convergence will be.