Robust data-induced learning control for uncertain nonlinear systems via dual-timescale adaptation.

Abstract

In addressing the challenge of achieving precise tracking control for uncertain nonlinear systems under complex disturbances, this paper proposes a Robust Data-induced Learning Control (R-DiLC) framework. The R-DiLC introduces a dual-timescale adaptation mechanism that effectively compensates for both iteration-invariant and iteration-varying disturbances. Iteration-invariant disturbances are addressed through historical error-driven adaptation, while iteration-varying disturbances are mitigated using robust control terms. This dual-timescale approach leverages operational data from exploratory trials and actual operations to refine control policies online, progressively enhancing disturbance rejection. Theoretical analysis proves the boundedness of all signals and gives a finite tracking-error bound for fixed finite command-filter bandwidths, while the contribution of non-repetitive disturbances is summable in the iteration-domain energy analysis. Simulation results validate the framework's efficacy, demonstrating effective tracking error reduction across iterations for repeatable industrial applications.