Building scientific consensus on the crash safety performance of automated driving systems

The deployment of Automated Driving Systems (ADS) offers a potential solution to reduce injury burden on US roadways. Rigorous retrospective safety assessments quantifying their efficacy is one key component in building scientific consensus around ADS safety. Unlike prior technologies, ADS evaluation is uniquely challenging due to its control over the entire driving task across diverse scenarios, use-case specific driving exposure profiles, and the unprecedented volume of real-time data generated.This paper focuses on improvements in data sources, implementation, and reporting to enable more credible performance assessments. Current public ADS data, like NHTSA SGO reports, lack the necessary exposure metrics (like Vehicle Miles Traveled, or VMT) for rate computation. We recommend ADS developers release detailed VMT and supplemental crash data to enable rate computation and rigorous analysis. Greater data granularity includes exposure confounders, comprehensive crash outcome units, and the inclusion of relevant performance lenses. Robust research implementation is critical. Researchers must align ADS and baseline data to account for temporal and geographic differences and are encouraged to follow best practices like the RAVE Checklist. To mitigate the potential for bias, we advocate for transparency through proactively published benchmarks and the use of justified reporting windows. Finally, independent oversight—including peer review, data access for independent researchers, and cooperative study partnerships—is essential for ensuring unbiased evaluation. The substantial influx of ADS data necessitates that the research community builds the empirical evidence for achieving consensus. Rigorous, continuous safety impact evaluations are a key catalyst; if ADS effectiveness is widely recognized, faster adoption could lead to expedited harm reduction on public roadways.