| 丁镱明,冯宇,李永强.完全信息下的四足机器人多对一追捕问题研究[J].高技术通讯(中文),2026,36(3):298~306 |
| 完全信息下的四足机器人多对一追捕问题研究 |
| Research on multi-to-one pursuit problem of quadruped robots under complete information |
| |
| DOI：10. 3772 / j. issn. 1002 - 0470. 2026. 03. 008 |
| 中文关键词: 追逃问题； 连续随机博弈； 深度强化学习； 虚拟自博弈； 纳什均衡 |
| 英文关键词: pursuit-evasion problem, continuous random game, deep reinforcement learning, virtual self-game, Nash equilibrium |
| 基金项目: |
| 作者 | 单位 | | 丁镱明 | （浙江工业大学信息工程学院杭州 310023） | | 冯宇 | | | 李永强 | |
|
| 摘要点击次数: 33 |
| 全文下载次数: 32 |
| 中文摘要: |
| 追逃问题在对抗、合作以及搜查等领域具有广泛应用。本文研究在完全信息条件下的多对一追逃博弈问题。其中，所有参与者的位置信息是互相公开的。在此基础上，构建了一个连续随机博弈框架，并借助不动点定理，在该框架下证明追逃博弈的纳什均衡策略的存在。为此，结合虚拟自博弈思想，提出了一种基于深度强化学习的方法，用于求解追捕双方最优策略。在仿真中与其他传统追捕算法进行了对比，本文算法的追捕胜率可达90%。最后，通过实物围捕平台进行实验，验证了本文所提出方法的有效性和实用性。 |
| 英文摘要: |
| The pursuit-evasion problem has wide applications in areas such as confrontation, cooperation, and search. This paper studies the multi-to-one pursuit-evasion game problem under complete information conditions. In this scenario, the positional information of all participants is publicly available. Based on this, a continuous random game framework is constructed, and using the fixed-point theorem, the existence of Nash equilibrium strategies for the pursuit-evasion game is proven within this framework. To achieve this, a method based on deep reinforcement learning is proposed, combining the concept of virtual self-game, to solve for the optimal strategies of both pursuer and evader. Then, comparisons are made with other traditional pursuit algorithms through simulations, demonstrating that the pursuit success rate of the proposed algorithm can reach 90%. Finally, experiments are conducted using a physical capture platform to validate the effectiveness and practicality of the proposed approach. |
|
查看全文
查看/发表评论 下载PDF阅读器 |
| 关闭 |
|
|
|
