Reinforcement learning‐modelled iterative learning control for tracking

Abstract Iterative learning control (ILC) has achieved significant success in repetitive control tasks. However, traditional ILC methods often lack the necessary feedback and parameter identification mechanisms to effectively handle applications subject to random disturbances or noise. To address this limitation, this paper introduces a reinforcement learning‐based iterative learning control (RLILC) algorithm specifically designed for tracking tasks under stochastic disturbances. Unlike conventional ILC approaches that rely on precise deterministic models, RLILC is a data‐driven method with the least squares estimation to deal with the raw data. It utilizes dual neural networks—a Predictor (P‐network) and an ‐network—to directly learn system dynamics and optimal learning gains from noisy data. The efficacy of this RLILC method is demonstrated through simulation experiments involving two distinct desired trajectories corrupted by random noise. The simulation results highlight its advantages: RLILC achieves the smallest tracking error compared to the other three methods evaluated, although it requires the largest control input.