A cross-methodological review on the safety effects of partial vehicle automation

We present a cross-methodological review on the safety implications of Partially Automated Vehicles (PAV). We inventory theoretical considerations on possible safety effects, and evaluate results from computer simulations, driving simulators, Experimental Driving Studies (EDS), Naturalistic Driving Studies (NDS), and real-world crash data (censuses), to gain insight in detrimental and beneficial safety effects of partial automation. Theoretical concerns center around incompatibility between the requirement of constant supervision and the potential for increased distraction, reduced vigilance, mode confusion and dis- and misuse, alongside technological limitations such as system reliability outside the Operational Design Domain (ODD). Computer simulations show mixed results, and suggest that benefits of automation may emerge only at higher levels of automation. Simulator studies, EDS and NDS support concerns about a detrimental impact of automation on human drivers, although beneficial effects on Surrogate Safety Measures (SSM), reflecting smoother driver behavior are also reported. These results are uncertain, due to limited generalizability of simulator studies and SSM. Census data indicate reduced crash rates but do not reveal causality. Detrimental effects on the human driver are a consistent finding across several methodologies, but studies adopting methodologies with higher representativeness nuance these findings, showing net safety benefits. Although automation may mitigate some theoretical concerns by relaxing requirements on the human driver, alternative explanations relating to confounding factors, such as the presence of dedicated safety systems, cannot be excluded. The findings emphasize the need for real-world validations, refinement of Human–Machine Interfaces (HMIs), and strategies to address fleet heterogeneity. This review underscores the critical role of integrated evidence in shaping policies for safer adoption of PAV technologies. • We perform a cross-methodological review on safety of Partially Automated Vehicles (PAV). • Different methodologies highlight human factors concerns and crash rate benefits. • Converging evidence from different methodologies provides a holistic view on PAV safety. • The findings suggest net safety benefits for partial automation but raise concerns about causality.