Dissertation Defense

Exploiting Structure in Safety Control

Zexiang Liu
3316 EECS BuildingMap
Zexiang Liu Defense Photo

PASSCODE: 760956


In safety-critical systems like autonomous vehicles, power systems, and robotics, ensuring compliance with safety constraints is vital. While various safety control methods have been  proposed, many do not fully exploit structures in dynamics, controllers, and disturbances. This dissertation aims to improve scalability and reduce conservativeness in safety control by leveraging these structures.

The first part of the dissertation focuses on scalable safety control methods. We start by showing the exponential convergence of the inside-out algorithm, a method for approximating the maximal robust controlled invariant set (RCIS). We then develop one-shot methods for computing implicit RCIS for controllable systems, utilizing a periodic structure in control. Additionally, we present a scalable method for approximating the maximal RCIS for input-delayed systems, by leveraging structural properties in dynamics.

In the second part of the dissertation, we focus on reducing the conservativeness in safety control, by exploiting structures in disturbance. One such structure is preview on disturbance. To assess the incremental value of preview information in safety control, we introduce a metric called safety regret, and show the exponential convergence of this metric for linear systems. For two special classes of systems with preview, we present efficient methods for computing their maximal RCIS. Finally, we introduce a novel safety control framework called opportunistic safety control, which enables safe operation both within and outside the maximal RCIS by leveraging the structures in disturbance models.


Chair: Necmiye Ozay