关键词: 多智能体深度强化学习, 互信息, 显式通信, 信息瓶颈理论, 合作环境

Abstract:

Effective explicit communication among agents in a multi-agent system can increase their capacity for cooperation. However， existing communication strategies typically use the agents’ local observations as the communication content directly， and the communication objects are usually fixed with a certain topology structure. On the one hand， these strategies are difficult to adapt to changes in tasks and environments， which causes uncertainty in the communication process. On the other hand， the communication objects and contents lack focus， resulting in some resource waste and lower communication effectiveness. To address the issues above， this paper proposes an approach that integrates deep reinforcement learning and information theory to realize multi-agent adaptive communication mechanism. The approach uses a prior network to allow the agent to dynamically choose the object， then utilizes the constraints of mutual information and the information bottleneck theory to effectively filter redundant information. Finally， the agent summarizes its own and received information to extract more effective information. The method proposed is demonstrated to improve the stability and interaction efficiency of multi-agent systems compared to other methods through cooperative navigation and traffic junction environments.

Key words: multi-agent deep reinforcement learning, mutual information, explicit communication, information bottleneck, cooperation environment