航空兵器2025，Vol.32Issue(5)：54-63,10.DOI:10.12132/ISSN.1673-5048.2025.0075
融合强化学习与改进人工势场的无人机编队路径规划
UAV Formation Path Planning Based on Reinforcement Learning and Improved Artificial Potential Field
摘要
Abstract
Aiming at problems of low planning efficiency and path oscillatory encountered in 3D UAV formation path planning using traditional artificial potential field(APF)methods,this paper proposes a collaborative path plan-ning strategy integrating deep reinforcement learning with improved APF.Firstly,the global optimal path for the leader UAV is generated through a double deep Q-network(DDQN)and prioritized experience replay mechanism,to address the suboptimal path issues inherent in conventional APF approaches.Then,an improved APF framework is developed by implementing adaptive adjustment strategies for gravitational coefficients,repulsive coefficients,and step sizes,to effectively suppress path oscillations and to enhance trajectory smoothness and convergence efficiency for follower UA-Vs.Finally,the DDQN-generated path is employed as a virtual leader,combined with the improved APF to achieve collaborative obstacle avoidance and coordinated path planning for multi-UAV formations.Simulation results demon-strate that the proposed approach can successfully guide the UAV formation to the target destination while achieving both obstacle and collision avoidance.Each UAV exhibits an average path length of 114 m and an average path smooth-ness of 2.3(°)/m.Compared with conventional methods,the proposed approach significantly improves formation path convergence efficiency and smoothness,while balancing the global optimality in path planning with UAV formation co-ordination.关键词
无人机编队/路径规划/路径震荡/深度强化学习/人工势场法/自适应参数Key words
UAV formation/path planning/path oscillatory/deep reinforcement learning/artificial potential field method/adaptive parameter分类
武器工业引用本文复制引用
赵天隆,陈龙胜,张存富,许贝..融合强化学习与改进人工势场的无人机编队路径规划[J].航空兵器,2025,32(5):54-63,10.基金项目
江西省自然科学基金项目(20232ACB202007 （）
20242BAB25085) （）
航空科学基金项目(20230007056001) （20230007056001）
