Trends in Deep Reinforcement Learning for Distributed Coordination and Cooperation with Homogeneous Multi-Agents

Deep Reinforcement Learning (DRL) has gained traction as a potent method for improving the real-time sequential decision-making abilities of autonomous vehicles, thereby stimulating a wealth of research in this area. However, in complex environments where autonomous vehicles must interact with various agents like pedestrians and other vehicles, the necessity for a multi-agent approach becomes evident. While agent communication is fundamental for efficient decision-making, its integration into DRL can be challenging. To address this problem, this paper explores inter-agent information sharing by a Communication Network (CommNet) that allows agents to efficiently make collective decisions only based on observed information. In this paper, the benefits of employing CommNet in a variety of real-world applications are evidenced, particularly where agents of autonomous vehicles must engage in information exchange in dynamic environments. Overall, the importance and potential benefits of the proposed strategy for autonomous vehicles are underscored to bolster their decision-making prowess.