Design of Robust Decentralized Controllers via Assume-Guarantee Contracts

TL;DR

This paper develops a method for designing robust decentralized controllers for linear systems with coupled subsystems, using assume-guarantee contracts and semidefinite programming to handle nonconvex information structures.

Contribution

It introduces a novel approach to address nonconvex decentralized control problems by treating information-coupling states as disturbances within ellipsoidal contract sets, enabling joint optimization.

Findings

The proposed method effectively handles nonclassical information structures.

Joint optimization over control policies and contract sets is achieved via semidefinite programming.

The approach improves robustness and tractability in decentralized control design.

Abstract

We consider the decentralized control of a discrete-time time-varying linear system subject to additive disturbances and polyhedral constraints on the state and input trajectories. The underlying system is composed of a finite collection of dynamically coupled subsystems, where each subsystem is assumed to have a dedicated local controller. The decentralization of information is expressed according to sparsity constraints on the state measurements that each local controller has access to. We investigate the design of decentralized controllers that are affinely parameterized in their measurement history. For problems with partially nested information structures, the optimization over such restricted policy spaces is known to be convex. Convexity is not, however, guaranteed under more general (nonclassical) information structures, where the information available to one local controller can be affected by control actions that it cannot reconstruct. To address the nonconvexity that arises in such problems, we propose an approach to decentralized control design where such information-coupling states are effectively treated as disturbances whose trajectories are constrained to take values in ellipsoidal "contract" sets whose location, scale, and orientation are jointly optimized with the underlying affine decentralized control policy. We establish a structural condition on the space of allowable contracts that facilitates the joint optimization over the control policy and the contract set via semidefinite programming.