A path following controller for deep-sea mining vehicles considering slip control and random resistance based on improved deep deterministic policy gradient. (15th June 2023)

• Record Type: Journal Article
• Authors: Chen, Qihang; Yang, Jianmin; Mao, Jinghang; Liang, Zhixuan; Lu, Changyu; Sun, Pengfei
• Is Part Of: Ocean engineering. Volume 278(2023)
• Publication Date: 2023-06-15
• Subjects: Deep-sea mining vehicle -- Path following -- Improved deep deterministic policy gradient -- Slip control -- Deep reinforcement learning; Ocean engineering -- Periodicals; Ocean engineering; Periodicals; 620.4162
• DOI: 10.1016/j.oceaneng.2023.114069
• Languages: English
• ISSNs: 0029-8018

Abstract

Abstract: This study aimed to develop a deep-sea mining vehicle (DSMV) path-following controller that could better reflect the actual deep-sea mining conditions. First, the dynamic model of the DSMV was improved. By introducing a nonlinear slip-control model and random environmental noise resistance, the controlled plant was developed to be closer to the actual mining operation condition. Second, an improved deep deterministic policy gradient (IDDPG) algorithm was proposed. Compared to the standard DDPG algorithm, the improved algorithm reduces the overestimation of the Q value and enhances the ability of an agent to explore the global optimum. A warm-up stage was introduced to improve stability at the beginning of training and accelerate the convergence speed of training. Third, a general reward function was designed for this type of problem. Combined with the uncertainty of the improved model, the generalization ability and adaptability to the unknown environment of the controller could be improved. Finally, through a random one-point-following training test in the simulation environment and different path-following comparison tests, the path-following control ability of the controller was verified. Highlights: Improved deep-sea mining vehicle model includes slip-control and noise resistance for nonlinear dynamics. Proposed deep RL controller has continuous output and general reward functions for path following. Improved controller's path-following ability verified via straight-line, S-curve, and standard path-following tests comparison.