首页 >

基于深度强化学习的空战机动决策试验研究

章胜1,周攀2,何扬2,黄江涛2,刘刚2,唐骥罡2,贾怀智3,杜昕4   

  1. 1. 中国空气动力研究与发展中心计算空气动力研究所
    2. 中国空气动力研究与发展中心
    3. 西北工业大小航天学院
    4. 中国空气动力研究与发展中心高速所
  • 收稿日期:2022-10-08 修回日期:2023-02-22 出版日期:2023-02-24 发布日期:2023-02-24
  • 通讯作者: 黄江涛
  • 基金资助:
    国家自然科学基金

Research on Air Combat Maneuver Decision-Making Flight Test Based on Deep Reinforcement Learning

  • Received:2022-10-08 Revised:2023-02-22 Online:2023-02-24 Published:2023-02-24

摘要: 空战智能决策将极大地改变未来战争的形态与模式。深度强化学习决策机可以挖掘飞行器潜力,是实现空战智能决策的重要技术范式,但是其工程实现鲜有报道。本文针对基于深度强化学习的双机近距空战机动智能决策的工程实现问题,开发了适于应用的深度神经网络在线机动决策模型,发展了通过飞行控制律跟踪航迹导引决策指令的机动控制方案,并进一步开展了软硬件实现工作与人机对抗飞行试验,实现了智能空战从虚拟仿真到真实飞行的迁移。研究结果表明基于本文发展的近距空战机动决策及控制方法,智能无人机在与人类“飞行员”的对抗中能够迅速作出有利于己方的动作决策,通过机动快速占据态势优势。本文工作显示了深度神经网络智能决策技术在空战决策中的潜在应用价值。

关键词: 近距空战, 智能决策, 深度强化学习, 人机对抗, 飞行试验

Abstract: The air combat intelligent decision-making will greatly change the form of wars. Deep reinforcement learning pro-vides a promising way of decision-making to explore and exploit the potential of unmanned aircrafts. It is an important technical paradigm to realize the intelligent decision-making in air combat. However, reports on its engineering implementation are rare. Aiming at the practical implementation of the maneuver intelligent decision-making based on deep reinforcement learning in the one-to-one fighters' close-range air combat, an online deep neural network decision-making model suitable for application is developed. The control scheme that tracks the transformed guidance commands with the flight control law is proposed. The corresponding software and hard-ware architecture are realized and the human-machine combat flight test is carried out, which realizes the transfer from the virtual simulation to the real flight in intelligent air combat. It is shown that based on the maneuver decision-making and control concept developed in this paper, the intelligent unmanned aircraft makes logical maneuver decisions and it occupies the advantageous situation quickly by maneuver when combatting the human "pilots". The flight test results demonstrate the potential application of the deep neural network intelligent decision-making machine in the air combat.

Key words: Close-range air combat, Intelligent decision-making, Deep reinforcement learning, Human-machine combat, Flight test

中图分类号: