APPLICATIONS OF GRAPHS IN MULTI-AGENT REINFORCEMENT LEARNING: RESEARCH PROGRESS AND METHODOLOGICAL FRAMEWORK

Zhang Che, Terekhov V. I., Afanasyev G. I.

Graph-based modeling is one of the key paradigms in multi-agent reinforcement learning (MARL), offering innovative solutions to complex problems in collaborative decision making. This paper explores the application of graph theory in MARL by providing a holistic methodological framework in three key areas: modeling of graph-structured environments, coordinated dynamic graph optimization, and graph-driven communication mechanisms. Based on this, a detailed comparison and analysis of state-of-the-art graph-based modeling methods in multi-agent reinforcement learning is provided. Case studies in traffic signal control and autonomous vehicle coordination demonstrate the versatility of the graph model in real-world applications, including intelligent transportation systems. However, significant theoretical challenges remain, such as ensuring convergence in graph-based dynamic communication and efficiently processing heterogeneous graph data.

Keywords:multi-agent reinforcement learning, multi-agent system, graph neural networks, coordination graph, dynamic graph