Characterization of maximum hands-off control

TL;DR

This paper characterizes the necessary and sufficient conditions for maximum hands-off control problems using an $L_0$-norm, demonstrating how sparsity is achieved and contrasting it with $L_1$ relaxations.

Contribution

It provides the exact necessary and sufficient conditions for maximum hands-off control using an $L_0$-norm, advancing the understanding of sparse control solutions.

Findings

$L_0$ cost yields sparse control solutions.

$L_1$ relaxation can lead to non-sparse solutions in singular problems.

Numerical examples illustrate the difference in control sparsity.

Abstract

Maximum hands-off control aims to maximize the length of time over which zero actuator values are applied to a system when executing specified control tasks. To tackle such problems, recent literature has investigated optimal control problems which penalize the size of the support of the control function and thereby lead to desired sparsity properties. This article gives the exact set of necessary conditions for a maximum hands-off optimal control problem using an $L_0$-(semi)norm, and also provides sufficient conditions for the optimality of such controls. Numerical example illustrates that adopting an $L_0$ cost leads to a sparse control, whereas an $L_1$-relaxation in singular problems leads to a non-sparse solution.