Data-driven Predictive Control for a Class of Uncertain Control-Affine Systems

TL;DR

This paper develops a distributionally robust, data-driven predictive control method for uncertain control-affine systems, providing high-probability performance guarantees and demonstrating effectiveness in traffic speed-limit control.

Contribution

It introduces a tractable reformulation of a stochastic optimization problem using distributionally robust optimization for uncertain systems, with an efficient online algorithm.

Findings

Effective control solutions prevent traffic congestion with high probability.

The approach offers high-probability performance guarantees.

Numerical simulations validate the control method's effectiveness.

Abstract

This paper studies a data-driven predictive control for a class of control-affine systems which is subject to uncertainty. With the accessibility to finite sample measurements of the uncertain variables, we aim to find controls which are feasible and provide superior performance guarantees with high probability. This results into the formulation of a stochastic optimization problem (P), which is intractable due to the unknown distribution of the uncertainty variables. By developing a distributionally robust optimization framework, we present an equivalent and yet tractable reformulation of (P). Further, we propose an efficient algorithm that provides online suboptimal data-driven solutions and guarantees performance with high probability. To illustrate the effectiveness of the proposed approach, we consider a highway speed-limit control problem. We then develop a set of data-driven speed controls that allow us to prevent traffic congestion with high probability. Finally, we employ the resulting control method on a traffic simulator to illustrate the effectiveness of this approach numerically.