Engineering Application Progress of Multi-Agent Deep Reinforcement Learning

Multi-agent reinforcement learning (MARL) is an important theoretical branch in the field of machine learning. It integrates methodologies from disciplines such as deep neural networks, reinforcement learning, cybernetics, game theory, and cognitive science, making it suitable for solving various complex multi-agent sequential decision-making problems. The multi-agent deep reinforcement learning (MADRL) method, which combines MARL with deep neural networks, has been widely applied to solve various real-world problems such as traffic signal control, autonomous driving, strategic games, and large model construction, demonstrating broad application prospects. This paper provides an overview of the progress of MADRL in engineering applications, which is divided into three main sections. Firstly, it reviews the research history and application algorithms of deep reinforcement learning (DRL), and outlines the concepts, methods, and problem representations of MADRL. Secondly, from the perspectives of three engineering fields: intelligent game theory, wisdom transport, and unmanned swarm system, it explores the latest engineering application examples of MADRL. Finally, it summarizes the challenges and prospects of MADRL in future engineering applications.