Model-Free Swing-Up and Balance Control of a Rotary Inverted Pendulum using the TD3 Algorithm: Simulation and Experiments

Le résumé fourni par la source

The Rotary Inverted Pendulum (RIP) system is a highly nonlinear and under-actuated mechanical system, which presents significant challenges for traditional control techniques. In recent years, Reinforcement Learning (RL) has emerged as a prominent nonlinear control technique, demonstrating efficacy in regulating systems exhibiting intricate dynamics and pronounced nonlinearity. This research presents a novel approach to the swing-up and balance control of the RIP system, employing a RL algorithm, Twin Delayed (TD3) Deep Deterministic Policy Gradient (DDPG), obviating the necessity for a predefined mathematical model. The physical model of the RIP was designed in SolidWorks and subsequently transferred to MATLAB Simscape and Simulink for the purpose of training the RL agent. The system was successfully trained to perform both swing-up and balance control using a single algorithm for both tasks, representing a significant innovation that eliminates the need for two or more separate algorithms. Additionally, the trained agent was successfully deployed onto an experimental model, with the results demonstrating the feasibility and effectiveness of the model-free TD3 approach in controlling under-actuated mechanical systems with complex dynamics, such as the RIP. Furthermore, the results highlight the sim-to-real transfer capability of this method.

Ce résumé expose les affirmations des auteurs. BNTIC ne l’interprète pas comme une validation indépendante des résultats.