Standards relevant to automated driving system safety: A systematic assessment

Automated Driving Systems (ADSs) in road vehicles, which can undertake the dynamic driving task without requiring a human driver, are on the verge of large-scale deployment. Assurance frameworks for vehicle systems have traditionally been underpinned by standards. Some commonly adopted standards will provide barriers to the market deployment of ADSs, particularly the inherent capability of ADSs to have their driving functionality changed using Machine Learning (ML) during in-service operation. New standards specific to ADSs are being adopted, but some of these also present challenges for ML-enabled system changes. Then there are emerging cross-sector AI and ML standards that will have implications for the automotive industry and regulators.This paper summarises an in-depth assessment of standards that are anticipated to play an important role in the safety assurance of ADSs. Summarised findings are provided, including key themes, covering issues such as the use of ADS safety cases that are strengthened by adopting complementary standards, ensuring safety assurance models cover the whole ADS lifecycle, and the use of a systems engineering approach that treats safety as a dynamic control problem and appropriately integrates software engineering standards. Other themes were more specific to ML, including the need for rigorous change processes, horizonal AI and ML standards to be appropriately considered by automotive manufacturers and regulators, and authorities to be clear on whether online changes that can occur autonomously without human oversight are allowed or not. The findings, themes and recommendations in this paper are intended to inform future safety assurance frameworks for ADSs