Crash Risk Assessment in a Mixed Traffic Environment of Autonomous and Human-Driven Vehicles

Abstract

This study investigates the interaction between autonomous vehicles (AVs) and human-driven vehicles (HDVs) during the transition phase of autonomous vehicle deployment. Utilizing two open-source datasets, Lyft L5 and Waymo, it analyzes car-following behaviors, revealing significant differences in space headway, time headway, and response times between AVs and HDVs. AVs typically maintain larger following gaps, which can impact traffic flow and safety dynamics in a mixed traffic environment. Furthermore, this research calibrates the Wiedemann 99 psychophysical car-following model to better replicate real-world driving characteristics and assess crash risks in such environments using dynamic Bayesian network. The results highlight that the implementation of AVs yields significant safety benefits – a remarkable 33.8% and 66.3% reduction in crash risk at 75% AV penetration, underscoring the positive impact of AVs in traffic safety improvement.