摘要
无论是在军事还是民用领域,无人机抛投应用价值都非常巨大。现有的无人机抛投方法有:基于人工经验的抛投、基于无人机运动动力学与空气动力学建模的抛投、基于导航系统的无人机抛投、基于视觉驱动的无人机自主精确抛投等。其中基于视觉驱动的无人机自主精确抛投方法涉及复杂的无人机与目标之间运动规律的精确感知与估计、空气动力与各种延时的影响,最佳抛投位姿和时机估计困难很大?本文拟通过基于深度学习的状态感知获取无人机的姿态角、晃动速度、地面目标相对关系等状态参数,通过强化学习解决视频采集处理传输延时、决策控制延时的影响,从而达到自主地、高精度地抛投?在此基础上设计和搭建了一个模拟实验平台,用于算法的训练和测试,并比较了基于视觉驱动的无人机自主精确抛投方法和依靠人工经验的抛投方法的效果,测试结果表明基于深度强化学习的视觉驱动无人机自主抛投方法获得相当高的成功率,能够有效的解决无人机精确抛投问题,该研究结果为无人机自主精确抛投提供了一种有效的解决方案和工程应用思路。
Whether in the military or civilian fields,the application value of UAV throwing is very huge.The existing UAV throwing methods include:throwing based on artificial experience,throwing based on UAV motion dynamics and aerodynamic modeling,UAV throwing based on navigation system,autonomous and accurate throwing UAV based on vision drive,etc.Among them,the autonomous and precise projection method of UAV based on vision drive involves the accurate perception and estimation of the motion law between the UAV and the target,the influence of aerodynamics and various time delays,and it is very difficult to estimate the optimal throwing position and timing.This paper intends to obtain the state parameters such as attitude angle,shaking speed,and relative relationship of ground targets of UAV through state perception based on deep learning,and solve the influence of video acquisition and processing transmission delay and decision control delay through reinforcement learning,so as to achieve autonomous and high-precision throwing.On this basis,a simulation experiment platform is designed and built for algorithm training and testing,and the effect of the autonomous and accurate projection method based on vision-driven UAV and the throwing method relying on manual experience is compared,and the test results show that the autonomous projection method of vision-driven UAV based on deep reinforcement learning has obtained a fairly high success rate and can effectively solve the problem of accurate UAV throwing,and the research results provide an effective solution and engineering application ideas for UAV autonomous and accurate throwing.
作者
陈易
吴昌峰
王新晴
CHEN Yi;WU Chang-Feng;WANG Xin-Qing(College of Field Engineering,Army Engineering University of PLA,Nanjing Jiangsu 210007,China)
出处
《机电产品开发与创新》
2023年第2期20-23,29,共5页
Development & Innovation of Machinery & Electrical Products
关键词
抛投
无人机状态感知
奖励反馈感知
目标检测
智能体
强化学习
Throwing
UAV state awareness
Reward feedback perception
Object detection
Agents
Reinforcement learning
