Sliding mode control for uncertain robot manipulators based on reinforcement learning

摘要

This paper investigates a new method that combines sliding mode control with reinforcement learning to improve the robustness and adaptability of robot manipulators. Sliding mode control is renowned for its effectiveness in managing disturbances and ensuring stability; However, it typically requires a precise understanding of system dynamics, which may not always be available. By combining forward learning and a super-twisting algorithm observer to estimate the state and uncertainty of the system, our framework enables the robot system to learn the optimal control strategy through interaction with the environment, thereby reducing dependence on precise model parameters. Through theoretical analysis and simulation experiments, we have demonstrated that the proposed hybrid control method not only maintains robust performance in the face of uncertainty, but also can dynamically adapt to constantly changing conditions. Extensive simulations were conducted on a two degree of freedom robot system to validate the effectiveness of the proposed method.