Iterative Learning Control (ILC) is a promising approach in industrial systems that perform repetitive tasks, but its application to Multi-Input-Multi-Output and Linear Parameter-Varying (LPV) systems is challenging due to model variations and mutual coupling effects. This study aimed to design transformation matrices that reduce dynamics variations, improving ILC convergence without relying on complex theoretical models. The proposed approach utilizes frequency response data to optimize transformation matrices, complemented by Gaussian process regression for real-time parameter tuning. This method reduces model variations, allowing the robustness filter in ILC to operate at higher cutoff frequencies without violating convergence conditions, improving tracking performance. This proposed approach was experimentally validated on a gantry stage system, demonstrating rapid settling.
Sasaki et al. (2026) studied this question.