Machine-Learned Specifications for the Verification and Validation of Autonomous Cyberphysical Systems

Abstract

Machine learning classifiers can be used as speci-fications for runtime monitoring (RM), which in turn supports evaluating autonomous systems during design-time and detecting/responding to exceptional situations during system operation. In this paper we describe how the use of machine-learned specifications enhances the effectiveness of RM for verification and validation (V & V) of autonomous cyberphysical systems (CPSs). In addition, we show that the development of machine-learned specifications has a predictable cost, at less than $100 per specification, using 2022 cloud computing pricing. Finally, a key benefit of our approach is that developing specifications by training ML models brings the task of developing robust specifications from the realm of doctoral-level experts into the domain of system developers and engineers.