Microsimulation and extreme value theory to assess the effect of varying levels of automated vehicles on crash risk

Abstract

Per the society of automotive engineers (SAE), automated vehicles (AVs) are likely to penetrate the system via five levels of automation. This implies that AVs and human-driven vehicles (HDVs) will coexist for a long time. The heterogeneity in traffic due to varying levels of AVs and HDVs would significantly affect safety because of uncertainty in vehicle-vehicle interactions. However, it is still unclear how the penetration of varying levels of AVs and HDVs will influence the crash risk. This study employed a microsimulation and extreme value theory (EVT) approach to assess the effect of varying levels of automated vehicles (AVs) on crash risk. A two-step process was proposed to calibrate the simulation model. In the first step, the simulation model was calibrated from an operational perspective, whereas the model was calibrated from a safety perspective in the second step. Different scenarios were formulated with different penetration rates of HDVs and varying levels of AVs. Traffic conflicts were extracted using the surrogate safety assessment model (SSAM), and the non-stationary peak-over threshold-based EVT approach was used to assess the crash risk. Relative crash risk (RCR) was used to evaluate the effect of varying levels of AVs on crash risk. The RCR compares the crash risk for different scenarios of AVs with the base scenario. For most scenarios, the RCR value was less than 1, highlighting improved safety levels under AV conditions. Further, the crash risk for both rear-end and lane-change conflicts was observed to increase with an increase in the penetration of Level 1 AVs. On the contrary, the crash risk decreased with an increase in the penetration of Level 2 and higher-level AVs. This highlights that higher safety benefits can be expected with increased penetration of Level 2 and higher-level AVs.