基于深度强化学习的人机智能对抗综述  

Survey of Human‑Computer Intelligence Gaming Based on Deep Reinforcement Learning

作　　者：刘玮 张永亮 程旭 LIU Wei;ZHANG Yongliang;CHENG Xu(College of Computer and Software,Nanjing University of Information Science and Technology,Nanjing 210044,China;Command and Control Engineering College,Army Engineering University of PLA,Nanjing 210007,China)

机构地区：[1]南京信息工程大学计算机与软件学院,南京210044 [2]陆军工程大学指挥控制工程学院,南京210007

出　　处：《指挥信息系统与技术》2023年第2期28-37,共10页Command Information System and Technology

基　　金：中国博士后科学基金(2018T111153)资助项目。

摘　　要：人机对抗是人工智能的热门领域,同时也为探索机器智能的内在原理与发展提供了途径。基于深度强化学习,讨论了人机智能对抗技术,并分析了人机对抗的内涵与机理。首先,简化了感知-判断-决策-行动(OODA)模型,总结了适用于深度强化学习的人机对抗框架,并归纳了态势认知、决策与优化以及协同与通信等关键技术;然后,阐述了态势特征提取与神经网络选择、策略制定与策略优化以及多智体训练模型与通信等技术内容;最后,列举了当前人机对抗的应用与挑战,并对人机对抗的未来发展做出了展望。The human-computer gaming is a hot field in artificial intelligence(AI),and then it also provides an effective way to explore the internal principle and development of machine intelligence.Based on the deep reinforcement learning,the technology of human-computer intelligent gaming is discussed,and the connotation and mechanism of human-computer gaming are analyzed.Firstly,the observation-orientation-decision-action(OODA)model is simplified,and the human-computer gaming framework for deep reinforcement learning is summarized.The key technologies including the situation awareness,the decision and optimization,and the collaboration and communication are summarized.Then,the contents of the situation awareness extraction and neural network selection technology,the strategy making and strategy optimization technology,and the multi agent training model and communication technology are described.Finally,the application and challenge of human-computer gaming are listed,and the future development of human-computer gaming is prospected.

关 键 词：人工智能 深度强化学习 人机对抗 认知决策 

分 类 号：TP18[自动化与计算机技术—控制理论与控制工程]