Expert data-induced learning control for uncertain nonlinear systems via statistical soft projection

This paper presents an Expert Data-Induced Learning Control (E-DiLC) framework, formulated within the Adaptive Iterative Learning Control paradigm, to achieve high-precision tracking for uncertain nonlinear systems in repeatable processes. While building upon the dual-timescale adaptation structure, the proposed method mainly contributes by systematically integrating statistical expert knowledge to constrain the learning process. Specifically, an expert knowledge-based smooth soft projection is designed to confine parameter estimates within physically feasible intervals determined from expert-provided data (expected values and confidence bounds). By restricting the adaptive search to a high-confidence region, the E-DiLC framework enhances robustness and accelerates convergence compared to conventional unguided adaptation. The global stability is formally proven via Lyapunov analysis, establishing error convergence and signal boundedness. The framework's effectiveness is demonstrated on a robotic manipulator, showcasing its ability to synergise expert knowledge with online adaptation for superior control performance.