Optimized backstepping tracking control using reinforcement learning for strict-feedback nonlinear systems with monotone tube performance boundaries
Gengning Zhang, Xin Wang, Ziming Wang, Ning Pang · International Journal of Control · 2025
Recently proposed monotone tube boundaries enhance performance but exacerbate the challenge of the entry capture problem (ECP). This article addresses the ECP under monotone tube boundaries and employs optimal control strategies combined with reinforcement learning (RL) to enhance performance in unknown state situations. By utilising error scaling functions (ESF), the proposed method effectively confines the error to a predefined neighbourhood in a specified time, without knowing the specific value of the initial error. Furthermore, this article innovatively incorporates the actor-critic structure into the monotone tube boundaries, significantly enhancing system performance. Due to the condition of state immeasurability, a neural network (NN)-based state observer is designed to estimate the states of the system. Additionally, command filters are used to simplify the multiple derivative calculations in traditional backstepping methods, addressing the ‘explosion of complexity’ problem. Mathematical derivations and simulation confirm its ability to meet predefined performance criteria.