Safety control, a quantitative approach ?

摘要

Safety control consists in maintaining the state of a given system inside a specified set of safe states. Traditionally, the problem is tackled using set-theoretic methods, which are mostly qualitative: states are partitioned between safety-controllable (i.e. states that belong to the maximal controlled invariant subset of the safe set) and safety-uncontrollable states. In this paper, we present a quantitative approach to safety controller synthesis. Our approach makes it possible to compute a measure of safety, which quantifies how far from the unsafe set (respectively, how close to the safe set) one can stay when starting from a given controllable (respectively, uncontrollable) state. For finite transition systems, such a measure can be computed in finite-time using a functional fixed-point iteration. In addition, we show that the level sets of the functional fixed-point coincide with the maximal controlled invariant subsets of a parameterized family of sets and that one can synthesize a common safety controller for all the sets of the family. In the second part of the paper, we show how the approach can be used in the framework of abstraction-based synthesis to lift these results to infinite transition systems with finite abstractions. To illustrate the effectiveness of the approach, we show an application of the approach to a simple boost DC-DC converter.