A co-design method of online learning SMC law via an input-mappping strategy

TL;DR

This paper introduces a novel co-design input-mapping sliding mode control method that leverages online learning from historical data to enhance convergence and performance in unknown dynamic systems.

Contribution

It proposes a co-designed sliding mode control framework using input-mapping and online learning, improving convergence without relying on traditional identification techniques.

Findings

Enhanced convergence rate demonstrated in simulations

Effective compensation for unknown system dynamics

Superior performance compared to traditional methods

Abstract

The research on sliding mode control strategy is generally based on the robust approach. The larger parameter space consideration will inevitably sacrifice part of the performance. Recently, the data-driven sliding mode control method attracts much attention and shows excellent benefits in the fact that data is introduced to compensate the controller. Nevertheless, most of the research on data-driven sliding mode control relied on identification techniques, which limits its online applications due to the special requirements of the data. In this paper, an input-mapping technique is inserted into the design framework of sliding mode control to compensate for the influence generated by the unknown dynamic of the system. The major novelty of the proposed input-mapping sliding mode control strategy lies in that the sliding mode surface and the sliding mode controller are co-designed through online learning from historical input-output data to minimize an objective function. Then, the convergence rate of the system is improved significantly based on the method designed in this work. Finally, some simulations are provided to show the effectiveness and superiority of the proposed methods.