The Quadratic-Quadratic Regulator Problem: Approximating feedback controls for quadratic-in-state nonlinear systems

TL;DR

This paper introduces an algorithm for approximating feedback controls in quadratic-in-state nonlinear systems, extending the linear-quadratic regulator to a more complex quadratic-quadratic regulator setting, and demonstrates its effectiveness on test problems.

Contribution

It presents a novel algorithm exploiting problem structure for the quadratic-quadratic regulator, enabling feedback control approximation for nonlinear systems with quadratic dynamics.

Findings

The algorithm effectively solves the QQR problem on test cases.

Modest improvements over linear feedback control laws were observed.

The method leverages tensor-based linear solvers for efficiency.

Abstract

Feedback control problems involving autonomous quadratic systems are prevalent, yet there are only a limited number of software tools available for approximating their solution due to the complexity of the problem. This paper represents a step forward in the special case where both the state equation and the control costs are quadratic. As it represents the natural extension of the linear-quadratic regulator (LQR) problem, we describe this setting as the quadratic-quadratic regulator (QQR) problem. This is significantly more challenging and holds the LQR as special case that must be solved along the way. We describe an algorithm that exploits the structure of the QQR problem that arises when implementing Al'Brekht's method. This approach is amenable to feedback laws with low degree polynomials but have a relatively modest model dimension that could be achieved by modern model reduction methods. This problem has an elegant formulation and a solution that introduces several linear systems where the structure suggests modern tensor-based linear solvers. We demonstrate this algorithm on a suite of random test problems then apply it to a distributed parameter control problem that fits the QQR framework. Comparisons to linear feedback control laws show a modest benefit using the QQR formulation.