Event-triggered reinforcement learning-based safe control for stochastic systems subject to asymmetric input constraints and unknown dynamics.

Abstract

This paper investigates the safe optimal control (SOC) for input-constrained unknown stochastic systems via adaptive dynamic programming (ADP) and generalized fuzzy hyperbolic model (GFHM). Firstly, a GFHM is employed to approximate the unknown nonlinear terms of the stochastic system, thereby eliminating the need for exact system modeling. Secondly, a modified nonquadratic value function is developed to address the problem of asymmetric input constraints (AICs), where a control barrier function (CBF) with a damping factor is incorporated to penalize unsafe behaviors. Thirdly, based on ADP, the optimal control policy is obtained by solving the modified Hamilton-Jacobi-Bellman (HJB) equation. Furthermore, a dynamic event-triggered mechanism (DETM) is proposed to alleviate the computational burden and conserve communication resources. In addition, the Lyapunov method is utilized to guarantee the uniform ultimate boundedness (UUB) of the system states and the weight estimation errors of the critic neural networks (CNNs). Finally, two simulation examples are provided to demonstrate the effectiveness of the proposed control strategy.