Research papers · 2022.07.08
Assurance Methodology for In-vehicle AI
Methodik zur Absicherung von KI im Fahrzeug
Overview
KI Absicherung researchers describe an iterative assurance method for DNN-based automotive perception. An ontology links the operating domain to data requirements and coverage analysis; development-time tests and runtime monitoring contribute evidence to a GSN safety argument. A pedestrian-detection example illustrates gaps in training data and targeted retraining.
Publication and version information
German ATZ, volume124, issue7–8, pages58–63. The English ATZ worldwide registry lists pages54–59; the project-hosted English copy prints58–63. The language editions are linked as one resource.