Adaptive Logarithmic Terminal Sliding-Mode Control With Deep Reinforcement Learning

Fast convergence, chattering reduction, and analytical derivation of the sliding time are key problems in sliding-mode control (SMC). In this article, a novel sliding surface, referred to as the logarithmic terminal sliding-mode (LnTSM) manifold, is proposed with an analytical solution for the sliding time. Then, a smooth controller is developed using the super-twisting algorithm and a barrier function technique. In the proposed scheme, the shared gain of two adaptive laws is dynamically tuned by a deep reinforcement learning-based agent to simultaneously ensure transient performance and steady-state precision. The stability of the proposed control scheme is proved via the Lyapunov theorem. Numerical simulations and experiments are subsequently provided to verify the above superiority. The results demonstrate that the proposed method exhibits significantly superior performance compared to other methods and has high value in practical applications.