Uncertainty Propagation in AI-driven Autonomous Systems

Modern autonomous systems rely on probabilistic AI techniques, complicating safety analysis and performance boundaries. This paper formalizes the relationship between Operational Design Domain (ODD) and uncertainty propagation in AI-driven systems. We propose a quantitative model for uncertainty propagation across series and parallel AI modules, enabling sub-module specification without re-evaluation. The approach is validated on real-world datasets and demonstrated through a case study in autonomous driving system, highlighting its effectiveness in modular specification and safety calibration.