计算机工程与应用 ›› 2020, Vol. 56 ›› Issue (5): 13-24.DOI: 10.3778/j.issn.1002-8331.1912-0100

• 热点与综述 • 上一篇    下一篇

多智能体深度强化学习研究综述

孙彧,曹雷,陈希亮,徐志雄,赖俊   

  1. 1.陆军工程大学 指挥控制工程学院,南京 210007
    2.中国人民解放军31102部队
  • 出版日期:2020-03-01 发布日期:2020-03-06

Overview of Multi-Agent Deep Reinforcement Learning

SUN Yu, CAO Lei, CHEN Xiliang, XU Zhixiong, LAI Jun   

  1. 1.College of Command and Control Engineering, Army Engineering University of PLA, Nanjing 210007, China
    2.Unit 31102 of PLA, China
  • Online:2020-03-01 Published:2020-03-06

摘要:

多智能体深度强化学习是机器学习领域的一个新兴的研究热点和应用方向,涵盖众多算法、规则、框架,并广泛应用于自动驾驶、能源分配、编队控制、航迹规划、路由规划、社会难题等现实领域,具有极高的研究价值和意义。对多智能体深度强化学习的基本理论、发展历程进行简要的概念介绍;按照无关联型、通信规则型、互相合作型和建模学习型4种分类方式阐述了现有的经典算法;对多智能体深度强化学习算法的实际应用进行了综述,并简单罗列了多智能体深度强化学习的现有测试平台;总结了多智能体深度强化学习在理论、算法和应用方面面临的挑战和未来的发展方向。

关键词: 强化学习, 深度学习, 多智能体系统, 多智能体深度强化学习

Abstract:

Multi-agent deep reinforcement learning is an emerging research hotspot and application direction in the field of machine learning and artificial intelligence. It covers many algorithms, rules, and frameworks, and is widely used in autonomous driving, energy allocation, formation control, trajectory planning,routing planning and social dilemma, it has extremely high research value and significance. The paper first briefly introduces the basic theory and development history of multi-agent deep reinforcement learning, then elaborates the existing classic algorithms according to four classification:non-association type, communication rule based type, mutual cooperation type and modeling learning type, then summarizes the practical application of multi-agent deep reinforcement learning and briefly lists the existing test platforms. The paper finally summarizes the challenges and future directions in theory, algorithms and applications of multi-agent deep reinforcement learning.

Key words: reinforcement learning, deep learning, multi-agent system, multi-agent deep reinforcement learning