Scholarlink

Multiple unmanned aerial vehicles (UAVs) pursuit-evasion game is a prominent approach for achieving air superiority. In this paper, a multi-agent independent soft actor-critic (MAISAC) and multi-agent independent decision transformer (MAIDT)-based offline reinforcement learning training strategy (MMOTS) is proposed to train multiple UAVs to complete the pursuit-evasion game, encompassing a two-stage design. In the first stage, we develop the MAISAC algorithm to facilitate policy improvement and offline dataset generation. In the second stage, MAIDT is proposed to realize the model training and pursuit-evasion tasks. Finally, simulation results demonstrate that the proposed MMOTS can achieve pursuit success rate exceeding 70% when the target distance ranges from 6 to 8, and the number of pursuit UAVs ranges from two to four. These outcomes underscore the outstanding generality, scalability and performance of MMOTS in the context of multi-UAV pursuit-evasion games. To accelerate relevant research in this direction, the code for simulation will be released as open-source.

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

MMOTS: A Multi-UAV Pursuit-Evasion Game Training Strategy Relying on Offline Reinforcement Learning

  • 会议论文

  • 发表日期：

    2026年

查看全文
  • Zekai Zhang,
  • Xiangjin Li,
  • Ziyu Chen,
  • Kangxin Hu,
  • Miao Peng,
  • Jingjing Wang,
  • Yong Ren

摘要

Multiple unmanned aerial vehicles (UAVs) pursuit-evasion game is a prominent approach for achieving air superiority. In this paper, a multi-agent independent soft actor-critic (MAISAC) and multi-agent independent decision transformer (MAIDT)-based offline reinforcement learning training strategy (MMOTS) is proposed to train multiple UAVs to complete the pursuit-evasion game, encompassing a two-stage design. In the first stage, we develop the MAISAC algorithm to facilitate policy improvement and offline dataset generation. In the second stage, MAIDT is proposed to realize the model training and pursuit-evasion tasks. Finally, simulation results demonstrate that the proposed MMOTS can achieve pursuit success rate exceeding 70% when the target distance ranges from 6 to 8, and the number of pursuit UAVs ranges from two to four. These outcomes underscore the outstanding generality, scalability and performance of MMOTS in the context of multi-UAV pursuit-evasion games. To accelerate relevant research in this direction, the code for simulation will be released as open-source.