海军航空大学学报2025，Vol.40Issue(4)：567-575,586,10.DOI:10.7682/j.issn.2097-1427.2025.04.007
基于分层强化学习的多无人机协同围捕方法
Multi-UAV Cooperative Encirclement Method Based on Hierarchical Reinforcement Learning
摘要
Abstract
Due to the complexity of multi-agent dimensions and the dynamics of the pursuit process,UAV encirclement often faces issues such as unclear division of labor and prolonged encirclement time.To address these issues,an option-based hierarchical reinforcement learning approach is introduced to the problem of multi-UAV cooperative encirclement.By using a multi-agent division of labor mechanism,the encirclement is divided into two stages:an upper-stage solving target allocation problem and a lower-stage solving path planning problem.Both stages adopt independent learning strate-gies,while achieving better control effects through a novel reward function.In terms of motion control,a maximum en-tropy reinforcement learning framework is introduced,using the Soft Actor-Critic(SAC)algorithm as the basic learning algorithm.Finally,the effectiveness of the proposed method is verified through simulation experiments.The results dem-onstrate that the method has good convergence and training efficiency.Compared with traditional methods,the upper-level controller using DRL can significantly reduce encirclement time and ensures a higher success rate.关键词
多无人机/协同围捕/围捕点分配/分层强化学习/围捕时间Key words
multi-UAV/cooperative encirclement/hunting point assignment/hierarchical Reinforcement Learning/encir-cling time分类
信息技术与安全科学引用本文复制引用
杨梓豪,王庆领..基于分层强化学习的多无人机协同围捕方法[J].海军航空大学学报,2025,40(4):567-575,586,10.基金项目
国家自然科学基金(62373102) （62373102）
江苏省自然科学基金(BK20221455) （BK20221455）
安徽省重点研究与开发计划(2022i01020013) （2022i01020013）
