Swing-Up and Stabilization Control of a Cart-Pendulum System Using NMPC, LQR, and Reinforcement Learning

The control of the cart-pendulum system is a challenging problem due to its nonlinear and unstable dynamics. This study evaluates three control strategies: Nonlinear Model Predictive Control (NMPC) + Linear Quadratic Regulator (LQR), Reinforcement Learning (RL), and a hybrid RL + LQR approach. Their performance is assessed regarding swing-up time, stabilization, and robustness. Results demonstrate that each control strategy presents distinct advantages and limitations, emphasizing the importance of selecting an appropriate approach based on application requirements. Furthermore, tuning the parameters of RL plays a crucial role in enhancing the efficiency and adaptability of these methods.