Reinforcement Learning-based Active Disturbance Rejection Control for Nonlinear Systems with Disturbance

This paper proposes a reinforcement learning-based active disturbance rejection controller (RL-ADRC) for trajectory tracking control of partially unknown nonlinear systems with external disturbances. It is also a complementary combination of RL and ADRC. In this method, an actor-critic-based RL algorithm is employed to explore an optimal and adaptive control strategy. Unlike traditional ADRC, which requires manual tuning of controller parameters, RL-ADRC utilizes an actor-critic network to approximate the optimal control strategy. By adopting the disturbance compensation philosophy from ADRC and integrating it with RL, RL-ADRC exhibits improved robustness against tracking errors, and the learning process is more efficient in terms of time.