Constrained Multi-Modal Density Control of Linear Systems via Covariance Steering Theory

TL;DR

This paper develops a method for controlling linear systems to steer their state distributions, modeled as Gaussian Mixture Models, towards desired distributions using covariance steering and optimization techniques.

Contribution

It introduces a novel control framework for density steering of linear systems with GMMs, employing randomized policies and optimization reformulations.

Findings

Effective control of GMM-based state distributions demonstrated

Optimization approach successfully achieves desired distribution matching

Bounds on distribution approximation errors established

Abstract

In this paper, we investigate finite-horizon optimal density steering problems for discrete-time stochastic linear dynamical systems whose state probability densities can be represented as Gaussian Mixture Models (GMMs). Our goal is to compute optimal controllers that can ensure that the terminal state distribution will match the desired distribution exactly (hard-constrained version) or closely (soft-constrained version) where in the latter case we employ a Wasserstein like metric that can measure the distance between different GMMs. Our approach relies on a class of randomized control policies which allow us to reformulate the proposed density steering problems as finite-dimensional optimization problems, and in particular, linear and bilinear programs. Additionally, we explore more general density steering problems based on the approximation of general distributions by GMMs and characterize bounds for the error between the terminal distribution under our policy and the approximated GMM terminal state distribution. Finally, we demonstrate the effectiveness of our approach through non-trivial numerical experiments.