Title: A Prototypical Expert-Driven Approach Towards Capability-Based Monitoring of Automated Driving Systems

Abstract: Supervising the safe operation of automated vehicles is a key requirement in order to unleash their full potential in future transportation systems. In particular, previous publications have argued that SAE Level 4 vehicles should be aware of their capabilities at runtime to make appropriate behavioral decisions. In this paper, we present a framework that enables the implementation of an online capability monitor. We derive a graphical system model that captures the relationships between the quality of system elements across different architectural views. In an expert-driven approach, we parameterize Bayesian Networks based on this structure using Fuzzy Logic. Using the online monitor, we infer the quality of the system's capabilities based on technical measurements acquired at runtime. Our approach is demonstrated in the context of the UNICAR.agil research project in an urban example scenario.