无人机集群不完全信息路径规划方法  

Incomplete information path planning method for an UAV cluster

作　　者：杜江涛 于家明 齐辉[1] DU Jiangtao;YU Jiaming;QI Hui(College of Aerospace and Civil Engineering,Harbin Engineering University,Harbin 150001,China;AVIC Shenyang Aircraft Design and Research Institute,Shenyang 110035,China)

机构地区：[1]哈尔滨工程大学航天与建筑工程学院,黑龙江哈尔滨150001 [2]航空工业沈阳飞机设计研究所,辽宁沈阳110035

出　　处：《哈尔滨工程大学学报》2024年第11期2210-2217,共8页Journal of Harbin Engineering University

基　　金：上海航天科技创新基金项目(SAST2020119).

摘　　要：针对不完全信息下路径规划的环境复杂与动态目标等问题,本文研究了无人机集群的路径规划策略。不完全信息下的无人机集群执行任务时,由于各无人机存在目标、始发地、时间等因素的差异,会有碰撞、冲突等问题。通过分析无人机动态战场与作战信息的不确定性,建立不完全信息的多无人机路径规划模型;基于多智能体深度确定性策略梯度算法设计无人机集群在探测时的全局奖励与部分局部奖励函数,使得训练后的无人机集群能够有效地在不完全信息的情景下,快速实现路径规划。仿真对比了无人机集群在不同参数、不同算法等条件下的学习效果,验证了改进多智能体深度确定性策略梯度算法在路径规划任务中的优越性。Aiming at the complex environment and dynamic target of path planning under incomplete information,this paper studies the path planning strategy of an unmanned aerial vehicle(UAV)cluster.When UAVs perform tasks under incomplete information,problems such as collision and conflict arise because of the difference in target,starting place,time and other factors.First,a multi-UAV path planning model with incomplete information is established by analyzing the uncertainty of UAV dynamic battlefield and operational information.Second,the multiagent deep deterministic policy gradient(MADDPG)algorithm is used to design the global reward function and partial local reward function of a UAV cluster during detection so that the trained UAV cluster can effectively and rapidly realize path planning under incomplete information.Finally,the simulation compares the learning effect of a UAV cluster under different parameters and algorithms and verifies the superiority of the improved MADDPG algorithm in path-planning tasks.

关 键 词：无人机集群 路径规划 不完全信息 动态目标 多智能体深度确定性策略梯度算法 强化学习 避障 仿真 

分 类 号：V249.122[航空宇航科学与技术—飞行器设计]