Fusion of reinforcement learning and analytical controller design theory for a class of nonlinear

Abstract

In this paper, we propose to combine the reinforcement learning method with the integral sliding mode control ISMC) for a class of nonlinear systems, input affine systems. A controller based on ISMC, which is firstly designed for a nominal linear model obtained by identification experiment etc., is used to make it easy for the reinforcement learning controller to learn especially in the early stage for an unstable plant and also to robustify the control system. To demonstrate the basic effectiveness of the proposed design method, an inverted pendulum driven by a cart is employed together with its numerical simulation.