Reinforcement learning-based saturated adaptive robust neural-network control of underactuated autonomous underwater vehicles
作者：
Highlights：
• A new robust bounded Actor–Critic neural network controller is designed for AUVs.
• The proposed reinforcement learning plan needn't knowledge about system dynamics.
• A new critic function is proposed that only depends on measurable system signals.
• The offered reinforcement learning plan is free of Hamilton–Jacobi–Bellman rule.
• The risk of actuator saturation is reduced by compensating its nonlinearity.
摘要
•A new robust bounded Actor–Critic neural network controller is designed for AUVs.•The proposed reinforcement learning plan needn't knowledge about system dynamics.•A new critic function is proposed that only depends on measurable system signals.•The offered reinforcement learning plan is free of Hamilton–Jacobi–Bellman rule.•The risk of actuator saturation is reduced by compensating its nonlinearity.
论文关键词：Adaptive robust control,Reinforcement learning,Actor–Critic neural network,Actuator saturation,Actuator nonlinearity
论文评审过程：Received 26 August 2020, Revised 22 October 2021, Accepted 20 February 2022, Available online 4 March 2022, Version of Record 9 March 2022.
论文官网地址：https://doi.org/10.1016/j.eswa.2022.116714
