Assessing the safety effectiveness of advanced driver assistance systems

Advanced Driver Assistance Systems (ADAS) have been increasingly integrated into modern vehicles to enhance safety and reduce crash risk, however, a comprehensive evaluation of their safety effectiveness across various traffic conditions is still needed.

This study assesses the real-world safety impact of ADAS using data from 3,549 ADAS-equipped vehicles and 3.76 million non-ADAS vehicles involved in two-vehicle crashes covering 34 models and 41 brands from 2017 to 2023. These ADASs are associated with specific driving maneuvers, including lane-changing with Blind Spot Warning (BSW), backing with Rear Cross Traffic Alert (RCTA), and straight-ahead driving with Forward Collision Warning (FCW), and are assumed to be activated during their respective maneuvers. The Quasi-Induced Exposure (QIE) method is applied to compare ADAS-equipped and non-ADAS-equipped vehicles across various conditions, analyzing key variables such as crash outcome, road condition, weather pattern, and driver behavior.

Results show that most ADAS-equipped vehicles are effective in reducing crashes under various traffic conditions when compared to non-ADAS-equipped vehicles. Additionally, this research compared a series of Bayesian hierarchical models and selected the best-fitting Bayesian Hierarchical Random-Parameter Model (BHRPM) considering unobserved heterogeneity to investigate the determinants influencing the crash responsibility for ADAS-equipped vehicle. Results reveal that intersections, dark unlit conditions, and higher speeds increase the likelihood of ADAS-equipped vehicle responsibility in a crash, while daylight, rural roads, and lower speeds decrease it. The interaction between vehicle type and ADAS is particularly critical.

The analysis highlights both the strengths and limitations of current ADAS, providing a detailed examination of specific ADAS safety functions in the contexts where they are most commonly deployed.