Examining the Effects of Non-Infrastructure Variables on the Safety Performance of Mixed Environments with Different Automated Vehicles Market Penetration Rates at Signalized Intersections

Abstract

To accommodate autonomous vehicles (AVs) and make them perform at their capacity, the current roadway infrastructure should be upgraded. However, upgrading the entire infrastructure cannot be conducted at once since it is time-consuming and costly. Therefore, evaluating alternative solutions to enhance or maintain traffic safety is critical, especially when primarily employing AVs. Among the roadway network components, intersections are one of the critical locations since they are proportionally experiencing more crashes, and signalized ones result in the majority of the fatal intersection related crashes. Hence, this dissertation examines the safety effects of various non-infrastructure variables at a signalized intersection in mixed traffic environments. To this aim, different levels of signal cycle length, speed limit, and left-turn (LT) signal phasing were considered under seven AV market penetration rates (MPRs). In addition, the safety effects of AV size were analyzed. A micro-simulation program was employed to develop and run 3,850 simulation runs that were developed using a full factorial design. Eventually, the traffic safety of the scenarios was analyzed from various aspects, including 1) 15 longitudinal driving volatility measure, 2) 15 lateral driving volatility measure, and 3) three machine learning (ML) regression models using the percentage of jerks. The results showed that: 1) increasing the AV MPR improved the majority of the lateral volatility measures, 2) larger AV size is associated with higher longitudinal and lower lateral volatility measures, 3) each volatility model is different, and each model could be implemented according to the objective of a study, 4) the ML models for total jerks, AV jerk, and regular vehicles (RVs) jerk consistently indicated that higher speed limits and permitted LT phasing results in the lower percentage of jerks, 5) increasing the AV size reduces the percentage of AV jerks and meanwhile has a negative effect on the RV jerks, but the benefit is higher, and 6) decreasing the cycle length reduces the number of jerky driving maneuvers for AVs. Eventually, the total and rear-end conflicts using the surrogate safety assessment model (SSAM) were compared to the number of jerks and indicated they could not be used interchangeably.