Die Akzeptanz autonomer Fahrzeuge hängt von deren Sicherheit ab. Für die Entwicklung sicherheitskritischer KI-Funktionen im Bereich des automatisierten Fahrens leistet Safe AI Engineering einen entscheidenden Beitrag zur praxistauglichen Absicherung.
Standardized safety for AI in autonomous vehicles.
A research project to define consistent safety requirements.
Safe AI Engineering’s research closes the gap between concept and proof of safety through verification, validation (V&V) as well as monitoring of AI. For this purpose, existing standards such as ISO/PAS 8800, ISO/PAS 21448 (SOTIF), and ISO 26262, which set international standards for AI functions, are integrated. The methodology is being developed using a camera-based AI perception function for pedestrian detection and tested in three use cases of increasing complexity: from a static scene with one pedestrian to dynamic, realistic traffic situations.
The project fits seamlessly into the project landscape of the VDA flagship initiative for autonomous and connected driving and is part of the second generation of the KI Familie (a project family that focuses on AI topics in automotive application fields) – together with the research projects jbDATA and nxtAIM.
Motivation
The integration of artificial intelligence into safety-critical systems means that AI functions must be proven to be reliably safe. This proof extends to both the correct functionality of automated vehicles and the assurance of continuous monitoring during operation. The basis for this verification is a structured AI engineering methodology. This methodology is being developed in the project using real and synthetic driving data in collaboration with automotive manufacturers, suppliers, IT and technology companies, and research institutions along the entire automotive value chain. The various disciplines of AI engineering, such as data generation, machine learning design, validation, and evaluation – bundled into subprojects – are woven together to form a safety rope that symbolizes the safety argumentation throughout the entire life cycle of an AI function.
Core innovations
Safe AI Engineering develops new approaches for the safe use of AI in automated driving, focusing on:
Safety and AI integration
Linking safety requirements with safety principles that contribute to the overall safety argumentation. This is achieved through the AI engineering methodology, which ensures traceability of the approval-relevant safety argumentation building at system and component level.
Database and quality assurance
Methods for the normalization of high-quality training and validation data, including synthetic data.
AI evaluation and safety standards
Evaluation of quality metrics according to ISO 26262, ISO/PAS 8800 and SOTIF.
Explainable and robust AI
Transparent methodology to validate an AI based perception function through explainable AI attributes.
Monitoring and continuous improvement
Model verification through evidence-based offline and online monitoring during development and in operation.
Practical demonstration
Testing the technologies in realistic demonstrators.
Impact
Safe AI Engineering strengthens the industry’s innovative power and promotes the further development of new technologies. The project contributes significantly to the integration of AI components in automated vehicles and enables AI-driven innovations in sensor technology, actuator systems, robustness, reliability, and data fusion and processing. This lays the foundation for scaling automated mobility solutions.
Vehicle manufacturers and suppliers benefit in two ways: they find a proven methodology for testing their AI systems already during development and in approval-relevant argumentation modules for safety-critical AI functions in the vehicle. The implementation of Level 3 and higher automated functions in the vehicle, including regulatory approval, is thus within reach. Social acceptance is increasing and users are getting a roadworthy vehicle – with reliable and safe AI functions.
Outlook
Safe AI Engineering points to the future, not only ensuring the safety of AI in vehicles, but also setting a standard for the entire industry. It is an important step toward the safe and reliable integration of AI into automated vehicles and helps to establish a general sense of safety with regard to automated and autonomous mobility systems among manufacturers, authorities, and users alike. Other areas, such as robotics or automation technology, can also at least learn from the AI Engineering method and my even adopt its approaches.