基于多智能体强化学习的对抗博弈技术综述

A Review of Adversarial Game Techniques Based on Multi-Agent Reinforcement Learning

摘要:

多智能体对抗系统是多方博弈的复杂系统,近年来很多研究聚焦于用强化学习解决多智能体对抗博弈问题。本文从多智能体强化学习的角度对智能博弈对抗的算法进行综述。首先对多智能体强化学习及博弈论进行简要介绍；然后提出多智能体强化学习的四项关键技术难点,并梳理相关解决方法；最后归纳多智能体强化学习的前沿研究方向,总结出三项研究热点与挑战。综述为后续的研究打下基础,为使用多智能体强化学习解决博弈对抗问题提供思路。

Abstract:

Multi-agent adversarial systems are complex multi-perty game systems, and in recent years, many studies have focused on using reinforcement learning to solve multi-agent adversarial game problems. This article will review intelligent game adversarial algorithms from the perspective of multi-agent reinforcement learning. First, a brief introduction to multi-agent reinforcement learning and game theory is given; then, four key technical difficulties of multi-agent reinforcement learning areproposed, and related solutions are sorted out; finally, the frontier research direction of multi-agent reinforcement learning is summarized, and three research hotspots and challenges are concluded. This review lays a foundation for the subsequent research and provides ideas for solving the game antagonism problem by using multi-agent reinforcement learning.