Survey of Multi-Agent Strategy Based on Reinforcement Learning

摘要

There are many multi-agent systems in life, such as driving vehicles, playing football games, and even bees building their hives. These systems are cooperative or competitive among multiple agents to fufill a task. Compared with single agent reinforcement learning, multi-agent has a larger search space, perception of other agents, and system robustness. The main purpose of this paper is to provide a clear overview of current multi-agent reinforcement learning strategy training methods, and to review the latest progress in multi-agent reinforcement learning. Finally, intorduced the application prospects and development trends of multi-agent reinforcement learning, summarized the technology of collaboration or competition. At present, multi-agent reinforcement learning has gradually been applied in many fields, such as robot systems, human-machine games, and autonomous driving. In the future, it will be widely used in resource management, transportation systems, medical care, finance and other fields.