Safety assurance for automated driving systems that can adapt using machine learning: A qualitative interview study.

Area of Science:

• Automotive Engineering
• Artificial Intelligence Safety
• Regulatory Science

Background:

• Traditional safety assurance frameworks struggle with Automated Driving Systems (ADSs) that use Machine Learning (ML) for in-service adaptation.
• Existing frameworks were not designed for autonomous operation without human drivers or ML-driven functional modifications.

Purpose of the Study:

• To identify themes for developing a safety assurance framework for ML-enabled ADSs.
• To assess expert support and feasibility of safety assurance concepts for ADSs.
• To gather feedback from global regulatory and industry stakeholders on ADS safety.

Main Methods:

• Qualitative in-depth interviews with leading global experts in ADS safety.
• Analysis of interview data to identify key themes and expert opinions.
• Focus on regulatory and industry perspectives on safety assurance for adaptive ADSs.

Main Results:

• Ten themes emerged, supporting a whole-of-life safety assurance for ADSs.
• Strong consensus on requiring Safety Cases from developers and Safety Management Plans from operators.
• Support for in-service ML-enabled changes within pre-approved boundaries, with mixed views on human oversight.

Conclusions:

• Support exists for reforming current regulatory frameworks for ADSs rather than wholesale changes.
• Challenges identified in regulators' knowledge, capability, and capacity to manage adaptive systems.
• Further research is needed to inform regulatory decisions on ADS safety assurance.