基于PER-IDQN的多无人飞行器围捕研究

Research of multiple UAVs pursuit-evasion based on PER-IDQN

针对单无人飞行器能力有限,无法完成复杂军事场景下对敌方机动目标的压制等问题,提出了一种基于深度强化学习的多无人飞行器围捕方法。基于围捕任务,构建了栅格化场景,并对具体环境做出了说明。在深度强化学习的框架中,考虑到多无人飞行器对目标的靠近合围、安全避障等因素,对各飞行器状态变量、动作输出、奖励函数进行了针对性的设计。仿真结果表明:经过深度强化学习PER-IDQN算法的训练后,多无人飞行器能够完成自主协同决策和威胁规避,实现对机动目标的包围捕获。

Aiming at the problem of the limited capability of a single UAV and the inability to suppress enemy maneuvering targets in complex military scenarios,a novel approach of multiple UAVs pursuit-evasion based on deep reinforcement learning is proposed.Based on the pursuit-evasion game,a rasterized scene is constructed and the specific environment is explained.In the framework of deep reinforcement learning,factors such as the proximity of multiple UAVs to the target and the safe avoidance of obstacles are considered,and the state variables,action outputs,and reward functions of each UAV are designed pertinently.The simulation results show that the multiple UAVs trained based on deep reinforcement learning PER-IDQN algorithm can make cooperative decision and achieve threat avoidance,and complete the task of pursuing the maneuvering target.