Reinforcement-Learning-Based Active Disturbance Rejection Control of Piezoelectric Actuators

Piezoelectric actuators (PEAs) are crutial components in nano-positioning processes due to their high positioning precision. However, the application of PEAs is hindered by their intrinsic rate-dependent hysteretic nonlinearity. This paper proposes a reinforcement-learning-based active disturbance rejection control (RLBADRC) scheme for controling PEAs. Unlike conventional preexisting linear active disturbance rejection control (LADRC) for PEAs, control parameters are adaptively learned with Q learning in the present RLBADRC instead. Moreover, the tracking errors are regarded as control states, whereas different control parameters are regarded as control actions. By this means, a Q table is established offline and parameters are adaptively learned with the well-trained Q table model online. Comparative experiments are conducted to verify the effectiveness of the approach.