Ensuring Safe and Trustworthy AI in Automotive Systems
This course provides a comprehensive exploration of AI safety in the automotive domain, focusing on emerging standards, risks, and practical methods for ensuring trustworthy AI in road vehicles. Using vision-based adaptive emergency braking as a running example, participants will study the interplay between functional safety (ISO 26262), SOTIF (ISO 21448), AI-specific standards (ISO 8800, ISO 5469), and regulatory frameworks like the EU AI Act. The course covers AI safety classification, risk factors, and concept-phase activities such as safety analysis, requirements, and acceptance criteria. Special emphasis is placed on data lifecycle management, dataset quality and challenges, as well as verification and validation strategies across the AI lifecycle—from design to field monitoring. Finally, participants will learn to build strong assurance arguments and safety cases, drawing from UL4600, ISO 8800, and current best practices for AI in autonomous and assisted driving systems.
Course Topics:
- Explore AI-related safety standards: ISO 26262, ISO 21448, ISO 5083, ISO 8800, ISO 5469, ISO 42001, and the EU AI Act
- Understand responsibilities, overlaps, and terminology across AI and safety frameworks
- Learn AI safety classification schemes and safety properties for road vehicles
- Identify risks in AI systems: performance gaps, robustness, transparency, bias, explainability, and malicious attacks
- Conduct AI safety analysis, requirements definition, and acceptance criteria in the concept phase
- Manage data lifecycle within the V-model, with emphasis on dataset quality, coverage, and vision-only challenges
- Apply AI verification and validation techniques: augmentation, synthetic data, calibration, runtime monitoring, and field reevaluation
- Develop and present AI safety cases using UL4600, ISO 8800, and assurance argumentation methods
- Address hot topics such as the safety of end-to-end (E2E) AI solutions and transitioning from aftermarket to series production
Requirements
Software: Chrome browser.
Hardware: Computer with an Internet connection, working speakers, and microphone.
Prior knowledge: Students should have prior knowledge in standards: automotive ISO 26262 (Fusa) and ISO 21448 (SOTIF).
