AUV path following controlled by modified Deep Deterministic Policy Gradient. (15th August 2020)

• Record Type: Journal Article
• Authors: Sun, Yushan; Ran, Xiangrui; Zhang, Guocheng; Wang, Xiangbin; Xu, Hao
• Is Part Of: Ocean engineering. Volume 210(2020)
• Publication Date: 2020-08-15
• Subjects: AUV -- Path following control -- Deep deterministic policy gradient -- Sample pools -- Average motion critic; Ocean engineering -- Periodicals; Ocean engineering; Periodicals; 620.4162
• DOI: 10.1016/j.oceaneng.2020.107360
• Languages: English
• ISSNs: 0029-8018

Abstract

Abstract: This study proposes a Deep Deterministic Policy Gradient algorithm based on optimized sample pools and average motion critic network (OSAM-DDPG) to realize the path following control of autonomous underwater vehicles (AUVs). The ideas of optimizing the sampling mode and the evaluation of motion are proposed to improve the efficiency of algorithm. OSAM-DDPG is used to train the force-to-state mapping of an AUV's dynamical model to realize its control. In the simulation test, the OSAM-DDPG algorithm only needs some episodes to obtain the complete control strategy. Based on the experience gained from the training, the problems of various paths following in the interference environment can be addressed, and the results demonstrate that the effect of path following control based on OSAM-DDPG is better than S-plane. Highlights: A new controller based on reinforcement learning was proposed. The ideas of optimizing the sampling mode and the evaluation of motion were proposed to improve the efficiency of algorithm. A fuzzy AUV dynamical model was trained as the environment. The anti-interference capability was considered in the proposed controller.