A Unified Safety Protection and Extension Governor
Nan Li, Yutong Li, Ilya Kolmanovsky
TL;DR
The paper addresses constrained safety for discrete-time systems by unifying a Safety Protection mechanism with a Safety Extension capability, enabling indefinite safety when possible or maximal delay to violation otherwise. It achieves this with a single continuous-variables optimization that uses an exponentially weighted penalty to encode the unknown horizon of safe operation, and proves exact-penalty properties under suitable assumptions. For linear systems with convex constraints, the online problem reduces to a convex quadratic program, ensuring efficient real-time computation. The approach is demonstrated on an automotive adaptive cruise control scenario, showing both robust safety guarantees and recoverable extension when disturbances challenge invariance. Overall, the work provides a theoretically sound and practically efficient framework for constrained safety with recovery opportunities in autonomous systems.
Abstract
In this paper, we propose a supervisory control scheme that unifies the abilities of safety protection and safety extension. It produces a control that is able to keep the system safe indefinitely when such a control exists. When such a control does not exist due to abnormal system states, it optimizes the control to maximize the time before any safety violation, which translates into more time to seek recovery and/or mitigate any harm. We describe the scheme and develop an approach that integrates the two capabilities into a single constrained optimization problem with only continuous variables. For linear systems with convex constraints, the problem reduces to a convex quadratic program and is easy to solve. We illustrate the proposed safety supervisor with an automotive example.