Reinforcement Learning-tuned Active Disturbance Rejection Controller for Tracking Control of a 4-DoF Robot Manipulator

In recent years, the use of robotics has significantly advanced numerous fields, offering enhanced precision and efficiency in complex tasks without requiring human intervention. One field of robotics that has hence attained a great deal of attention and technological advancements is the control of robots. The Active Disturbance Rejection Control (ADRC) method, which can be considered as advanced iteration of the PID controller, stands out for its exceptional disturbance handling capabilities, garnering recognition in various industries. However, the manual adjustment of ADRC parameters proves time-consuming, prompting exploration into intelligent techniques for automatic parameter tuning. In this paper, we investigate Reinforcement Learning (RL) as a promising approach for optimizing parameters, specifically focusing on the tracking control of a 4 Degree of Freedom (DOF) robot manipulator. The obtained simulation results demonstrate the successful training of the system, showcasing its proficiency in tracking control tasks. By leveraging RL, we achieve efficient parameter optimization, reducing the need for manual intervention. This research not only highlights the potential of RL in enhancing ADRC performance but also underscores its broader applicability in advancing robotics integration with various fields for enhanced precision and automation.