摘要
随着深度强化学习技术的快速发展,将其应用于飞行器控制领域成为研究热点。针对深度强化学习方法在飞行器控制中的应用问题,概述了深度强化学习的演变历史和发展现状,介绍了深度强化学习的典型应用场景和基本原理。进一步介绍了两种面向飞行控制的算法训练平台,明确了不同网络结构的控制特性及由飞行状态构建控制网络输入数据的方法。分析了将深度强化学习方法应用于飞行器控制中存在的问题,提出了相应的解决方案,并对其未来发展方向进行了展望。
With the rapid development of deep reinforcement learning technology,its application in the field of aircraft control has become a research hotspot.In view of the application of deep reinforcement learning methods in aircraft control,the evolution history and development status of deep reinforcement learning are summarized,and the typical application scenarios and basic principles of deep reinforcement learning are introduced.It further introduces two flight control-oriented algorithm training platforms,and clarifies the control characteristics of different network structures and the method of constructing control network input data from flight status.The problems in applying deep reinforcement learning methods to aircraft control are analyzed,corresponding solutions are proposed,and the future development direction is prospected.
作者
甄岩
袁健全
池庆玺
郝明瑞
Zhen Yan;Yuan Jianquan;Chi Qingxi;Hao Mingrui(Science and Technology on Complex System Control and Intelligent Agent Cooperation Laboratory,Beijing 100074,China)
出处
《战术导弹技术》
北大核心
2020年第4期112-118,共7页
Tactical Missile Technology
关键词
飞行器控制
深度强化学习
值函数
策略梯度
训练平台
aircraft control
deep reinforcement learning
value function
strategy gradient
training platform