Scholarlink

<p>In this paper, an optimal three-dimensional (3D) path-following control scheme incorporating state estimation is proposed for the underactuated autonomous underwater vehicle (AUV) operating under sparse sensor attacks and measurement disturbances, which effectively addresses the underactuation of the optimal control problem. To reduce the impacts of sparse sensor attacks and measurement disturbances, a continuous-time unscented Kalman filter-based state estimation algorithm with attack data isolation is proposed, and a new adaptive measurement covariance matrix is designed in this paper. The novel algorithm not only isolates attack data and improves estimation accuracy under measurement disturbances, but also accounts for the randomness of attack moments. With the aim of reducing the waste of limited resources and considering the underactuation problem of the AUV, an improved optimal 3D path-following method incorporating historical estimated weights based on a reinforcement learning frame via neural networks is designed in this paper. All closed-loop signals of the AUV are proved to be semi-globally uniformly ultimately bounded through Lyapunov stability theorem. Finally, simulation results with comparisons show the validity of control scheme and estimation algorithm.</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

Reinforcement learning-based optimal 3D path-following control incorporating state estimation for AUV under sparse sensor attacks and measurement disturbances

  • 期刊论文

  • 发表日期：

    2025年10月9日

查看全文
  • Ziyi Qiu,
  • Zhechen Zhu,
  • Yingnan Pan,
  • Yan Lei

摘要

In this paper, an optimal three-dimensional (3D) path-following control scheme incorporating state estimation is proposed for the underactuated autonomous underwater vehicle (AUV) operating under sparse sensor attacks and measurement disturbances, which effectively addresses the underactuation of the optimal control problem. To reduce the impacts of sparse sensor attacks and measurement disturbances, a continuous-time unscented Kalman filter-based state estimation algorithm with attack data isolation is proposed, and a new adaptive measurement covariance matrix is designed in this paper. The novel algorithm not only isolates attack data and improves estimation accuracy under measurement disturbances, but also accounts for the randomness of attack moments. With the aim of reducing the waste of limited resources and considering the underactuation problem of the AUV, an improved optimal 3D path-following method incorporating historical estimated weights based on a reinforcement learning frame via neural networks is designed in this paper. All closed-loop signals of the AUV are proved to be semi-globally uniformly ultimately bounded through Lyapunov stability theorem. Finally, simulation results with comparisons show the validity of control scheme and estimation algorithm.