Q-Learning Based Control for Swing-Up and Balancing of Inverted Pendulum

Abstract

This study addresses the classic control problem of stabilizing an inverted pendulum on a moving cart, a challenge in control theory and robotics due to its inherent instability and highly nonlinear dynamics. We explore the application of Q-learning, a model-free reinforcement learning algorithm, and its efficacy in deriving an optimal control policy for the system without precise system models. Our approach utilizes Q-learning's capacity for stabilizing a pendulum in an upright position on the top of the horizontally moving cart within a certain boundary. Our strategy adapts to a dynamic environment while showcasing its robustness in developing control policies for complex systems. This research bridges classical control theory with reinforcement learning techniques, contributing to the domain by demonstrating the versatility and potential of machine learning in control tasks.