Mosaic: Model-based Safety Analysis Framework for AI-enabled Cyber-Physical Systems

• DOI: 10.48550/arxiv.2305.03882
• Publication Date: 2023-01-01
• Authors: Xuan Xie; Jiayang Song; Zhehua Zhou; Fuyuan Zhang; Lei Ma
• Keywords: Cyber-physical system

Cyber-physical systems (CPSs) are now widely deployed in many industrial domains, e.g., manufacturing and autonomous vehicles. To further enhance the capability applicability of CPSs, there comes a recent trend from both academia industry to utilize learning-based AI controllers for system control process, resulting an emerging class AI-enabled cyber-physical (AI-CPSs). Although such AI-CPSs could achieve obvious performance enhancement lens some key requirement indicators, due random exploration nature lack systematic explanations their behavior, AI-based techniques also bring uncertainties safety risks controlled system, posing urgent need effective analysis AI-CPSs. Hence this work, we propose Mosaic, model-based framework Mosaic first constructs Markov decision process (MDP) model as abstract AI-CPS, which tries characterize behaviors original AI-CPS. Then, based on derived model, is designed two aspects: online monitoring offline model-guided falsification. The usefulness evaluated diverse representative industry-level AI-CPSs, results demonstrate that providing enables outperform state-of-the-art falsification techniques, basis advanced