Adaptive robust tracking control with active learning for linear systems with ellipsoidal bounded uncertainties

TL;DR

This paper introduces an adaptive robust control method for linear systems with uncertainties bounded within ellipsoids, actively learning these bounds to improve tracking performance and reduce uncertainties efficiently.

Contribution

It proposes a novel active learning approach using recursive set-membership estimation to adaptively learn ellipsoidal uncertainty bounds in real-time.

Findings

Enhanced tracking accuracy compared to fixed-ellipsoid methods

Faster uncertainty reduction through active learning

Computationally efficient second-order cone programming solution

Abstract

This paper is concerned with the robust tracking control of linear uncertain systems, whose unknown system parameters and disturbances are bounded within ellipsoidal sets. We propose an adaptive robust control that can actively learn the ellipsoid sets. Particularly, the proposed approach utilizes the recursive set-membership state estimation in learning the ellipsoidal sets, aiming at mitigating uncertainties in the system control. Upon the learned sets representing the recognized uncertainties, we construct a robust control with one-step prediction for system output tracking. In deriving an optimized control law, we reformulate the optimization objective into a second-order cone programming problem that can be solved in a computationally friendly way. To further stimulate the active learning of uncertainties over the control procedures, we enrich the information used for the learning by maximizing the volume of the ellipsoid set, supposed to lead to increased learning accuracy and accelerated uncertainty reduction. To verify our approach, we conduct numerical simulations to compare the fixed-ellipsoidal-set robust control with ours, and investigate the positive effect of the designed active learning in the uncertain system control process.