Scholarlink

<p>Under the framework of deep reinforcement learning (RL), the task completion capability of underactuated underwater unmanned vehicle (UUV) in a continuous multi-task environment with unknown disturbances and actuator faults is investigated. Based on a novel forward-looking sonar, a new pathtracking motion control framework incorporating an obstacle avoidance task is proposed. On this basis, the D-TD3 algorithm was proposed, which realizes the collection of state classification and automatic task switching through multi-experience pooling and event triggering mechanism. In addition, a feed-forward compensation mechanism based on reduced-order extended state observer (RESO) is designed to compensate for the effects of unknown disturbances and actuator faults, which is embedded in our proposed algorithm. Finally, the effectiveness of the proposed method is verified through the simulation of path tracking and obstacle avoidance of the virtual UUV model in the continuous control task.</p>

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

An adaptive path tracking with obstacle avoidance scheme for UUVs under complex underwater conditions

  • 期刊论文

  • 发表日期：

    2025年7月23日

查看全文
  • Xun Zhang,
  • Ziqi Wang,
  • Zhiguang Feng,
  • Haiping Du

摘要

Under the framework of deep reinforcement learning (RL), the task completion capability of underactuated underwater unmanned vehicle (UUV) in a continuous multi-task environment with unknown disturbances and actuator faults is investigated. Based on a novel forward-looking sonar, a new pathtracking motion control framework incorporating an obstacle avoidance task is proposed. On this basis, the D-TD3 algorithm was proposed, which realizes the collection of state classification and automatic task switching through multi-experience pooling and event triggering mechanism. In addition, a feed-forward compensation mechanism based on reduced-order extended state observer (RESO) is designed to compensate for the effects of unknown disturbances and actuator faults, which is embedded in our proposed algorithm. Finally, the effectiveness of the proposed method is verified through the simulation of path tracking and obstacle avoidance of the virtual UUV model in the continuous control task.