A multi-agent planning approach integrated with learning mechanism

This paper presents a multi-agent planning approach integrated with learning mechanism. This method involves in task allocation, path planning, avoiding conflicts, cooperation, parameter learning, pattern learning, etc. In addition, with this method a multi-agent Sokoban platform is defined. With some simulations on this platform, the advantages of multi-agent planning approach with learning mechanism are illustrated comparing with single-agent approach. Since the proposed method can improve the efficiency and capability of multi-agent planning, by referring to the results of this research, the proposed method will be adopted for multi-robot system in the future research.