AUV position tracking and trajectory control based on fast-deployed deep reinforcement learning method. (1st February 2022)

• Record Type: Journal Article
• Authors: Fang, Yuan; Huang, Zhenwei; Pu, Jinyun; Zhang, Jinsong
• Is Part Of: Ocean engineering. Volume 245(2022)
• Publication Date: 2022-02-01
• Subjects: AUV -- Deep reinforcement learning -- Position tracking -- Trajectory control; Ocean engineering -- Periodicals; Ocean engineering; Periodicals; 620.4162
• DOI: 10.1016/j.oceaneng.2021.110452
• Languages: English
• ISSNs: 0029-8018

Abstract

Abstract: Aiming at the difficult problem of motion control of under-actuated and X-rudder autonomous underwater vehicle (AUV), the present work adopts deep reinforcement learning (DRL) method for its posture control. First, an AUV agent is trained with deep deterministic policy gradient (DDPG) algorithm in a simulation environment, and three-degree-of-freedom posture control of the AUV at a constant speed, fixed roll, variable pitch, and variable yaw, is successfully achieved. Subsequently, the AUV's yaw angle range is extended, and the control failure problem when AUV's yaw angle approaches a critical value is solved, realizing the rapid deployment of the DRL algorithm for AUV control. On this basis, the position-tracking task of AUV for targets in different orientations in three-dimensional space is completed, achieving a six-degree-of-freedom control of AUV. Additionally, by decomposing the trajectory control task of AUV in three-dimensional space into multiple position-tracking missions, the trajectory control of AUV in the underwater horizontal plane and underwater three-dimensional space is realized, demonstrating the significant task generalization ability of the control methods proposed. Highlights: The AUV position tracking and trajectory control is investigated with the effective application fast-deployed deep reinforcement learning method. The agent can be trained and the control method could be deployed rapidly. The proposed method could achieve attitude keeping, position tracking, and trajectory control missions, demonstrating its excellent generalization capability. By decomposing the trajectory control task into multiple position tracking missions, the trajectory control of AUV in three-dimensional space is effectively realized. This paper realizes the 6-degree-of-freedom AUV control to conduct multiple missions, which is more maneuverable and complicated than traditional cross-rudder AUV.