REINFORCEMENT LEARNING ASSISTED SLIDING MODE CONTROLLER FOR A ROTARY INVERTED PENDULUM

Abstract

A Rotary inverted Pendulum (RIP) is an underactuated system and it serves as a standard for non-linear control systems. Due to the system's extreme sensitivity to outside disturbances, uncertainties, and coupling issues between rotation arm and pendulum movement, controlling the inverted pendulum in vertical position poses difficult challenges. It has been found that SMC is an aggressive control approach, but it has certain drawbacks including chattering effect and parameter selection challenges. Reinforcement Learning, which makes use of an interaction between the agent and its environment to train and make decisions, has a lot of potential for non-linear control issues that do not require mathematical modelling. The RIP system is suggested to use a combination of Reinforcement Learning and Sliding Mode Control in this study. Reinforcement Learning is used to modify the parameters of SMC to obtain better stability and robustness. Simulations via MATLAB/Simulink are used to analyse the performance of the recommended controller.