The convergence of Generative AI (GenAI) and Functional Safety (FuSa) under ISO 26262 is reshaping the landscape of Advanced Driver Assistance Systems (ADAS) by integrating intelligence with reliability. GenAI revolutionizes perception, prediction, and decision-making in dynamic environments, enabling vehicles to navigate complex real-world scenarios with unprecedented adaptability. However, its probabilistic nature, lack of transparency, and vulnerability to edge-case failures pose significant challenges in ensuring compliance with functional safety standards.
This paper explores how GenAI enhances object detection, pedestrian prediction, and risk assessment, ultimately reducing accident rates and fostering safer autonomous operations. Furthermore, GenAI-driven synthetic data generation accelerates ADAS validation, allowing comprehensive testing of safety-critical scenarios beyond conventional datasets. However, integrating GenAI into safety-compliant systems requires overcoming determinism constraints, explainability concerns, and cybersecurity vulnerabilities such as adversarial attacks and biased decision-making.
To bridge this gap, we propose a hybrid approach combining deterministic safety mechanisms with GenAI-driven adaptability while leveraging Explainable AI (XAI) techniques to enhance transparency. Additionally, adherence to emerging standards like SOTIF (ISO 21448) and the development of rigorous verification frameworks are crucial for ensuring real-time, fail-safe operations.
By addressing these challenges through robust safety assurance, cross-disciplinary collaboration, and regulatory alignment, this research underscores GenAI’s transformative potential in ADAS while upholding the stringent reliability demands of ISO 26262. This paper presents a roadmap for realizing the next generation of safe, intelligent, and adaptive driving systems.