Review on Dec-POMDP Model for MARL Algorithms

Abstract

With the various application and rapid development of deep reinforcement learning in multi-agent systems, the field of multi-agent reinforcement learning (MARL) continues to emerge with various practical application problems and corresponding solutions. Since single-agent decision-making tasks are difficult to meet the demands of complex application scenarios, most tasks need to be accomplished through cooperation or competition among multiple agents. This paper first introduces the characteristics of multi-agent systems, the basic concepts of reinforcement learning, and the methods for modeling multi-agent systems, then analyzes the strengths and weaknesses of two multi-agent reinforcement learning models for MARL algorithms in chronological order, which are communication-based learning model and collaborative-based learning model. Finally, based on the emerging opportunities and challenges in the field of MARL, possible research directions are proposed for the researchers’ reference.KeywordsMARLDec-POMDP modelCooperative environment