多智能体协同研究进展综述: 博弈和控制交叉视角

Multi-agent collaboration is widely applied and considered a key component for breakthroughs in next-generation artificial intelligence (AI) foundational theories, possessing significant scientific and engineering value. As AI technology advances, the traditional single-control perspective on multi-agent collaboration is inadequate for large-scale complex tasks. Thus, the integration of game theory and control has emerged, offering greater flexibility, adaptability, and scalability, thereby expanding the potential of multi-agent systems. This paper begins by reviewing progress in the field of multi-agent coordination control and estimation. It then introduces the basic concepts of game theory, focusing on modeling and analysis of multi-agent collaboration within the framework of differential games, and briefly summarizes the use of reinforcement learning algorithms to solve game equilibria. Two typical multi-agent collaboration scenarios, namely multi-robot navigation and electric vehicle charging scheduling, are discussed to demonstrate how the integration of game theory and control addresses key challenges. Finally, the paper summarizes and provides an outlook on multi-agent collaboration within the integrated game-control framework.