Doctoral Defense - Md Tanvir Ashraf

About the Event

The Department of Civil and Environmental Engineering 

Michigan State University

Ph.D. Dissertation Defense

Thursday, June 26, 2025 at 1:00 PM EST

3546D Engineering Building

ABSTRACT

AUTONOMOUS VEHICLE BEHAVIOUR in MIXED TRAFFIC

By: Md Tanvir Ashraf

Advisor: Kakan Dey

Autonomous Vehicle (AV) technology has been maturing over the last decade through continuous testing on controlled test tracks, public roadways, and simulation environments. Understanding the driving behavior of AVs in mixed-traffic streams by analyzing their interactions with other human-driven vehicles (HDVs) is essential for ensuring AVs’ seamless integration into existing transportation systems. This Ph.D. dissertation utilized publicly available AV driving datasets from Waymo, Lyft, and Argoverse to investigate AVs’ car-following (CF), merging, and lateral crossing conflict resolution behaviors in a mixed traffic environment. The specific research objectives are to- (i) investigate CF behavior (e.g., headway, gap time) of AVs and identify the non-linear time series relationship in safety related conflict measures; (ii) analyze crossing conflicts at intersections for AV-involved and non-AV scenarios based on driving volatility measures; and (iii) study the merging behavior of AVs in mixed traffic environments.

AV’s rear-end collision risk for different CF scenarios in mixed traffic (i.e., AV as a follower or leader vehicle) was investigated to establish a relationship between CF conflict indicators and rear-end crash risks. The traditional Extreme Value Theory (EVT) and the Deep Learning (DL) methods were combined to perform traffic conflict-based surrogate safety analysis and real-time crash risk estimation. Relative crash risks between different CF pairs, estimated from the EVT parameters, indicate that CF scenarios involving an HDV following an AV had two to three times higher rear-end crash risk than those involving an HDV following another HDV. Empirical analysis of time-to-collision (TTC) and deceleration rate to avoid a collision (DRAC) measure revealed that AVs’ CF behavior was more cautious and maintained higher time gaps and headways than HDVs. When AV was the follower vehicle in a CF pair, the relative crash risk was below one compared to the HDV following HDV scenario, indicating that AVs reduced the rear-end crash risk when following HDVs.

Conflict scenarios were categorized into AV-involved and non-AV scenarios to investigate AVs’ crossing conflict resolution behaviors at intersections. AV-involved scenarios were further classified into AV-first and AV-second scenarios depending on whether AVs passed through the conflict region first or second. Using a hierarchical clustering algorithm, AVs' crossing conflict resolution behavior was then categorized into high, medium, and low risk driving categories based on driving volatility measures. Approximately 29% of the conflicts in the AV-first scenario (in which HDV was the following vehicle) exhibited a high crossing conflict risk. All AV-second scenario conflicts were classified as low-risk or medium-risk. The Bayesian hierarchical model results indicated AVs had safer interactions with other roadway users (i.e., HDVs, pedestrians, and cyclists) while maintaining higher speeds and uniform driving profiles. The interaction of AVs with vulnerable road users (i.e., pedestrians and cyclists) demonstrated a lower crash risk compared to HDVs, indicating AVs’ safer driving behavior. Additionally, AVs exhibited safer conflict resolution behavior in performing unprotected left turns at the intersection than HDVs.

AVs’ and HDVs’ merging event data showed that, contrary to CF and crossing conflict behaviors, the merging behaviors of AVs and HDVs were similar in a mixed traffic environment, in terms of gap time (GT). Higher GT reduced the following vehicle’s speed variation at the target lane for AV and HDV merging events. A Weibull random parameter hazard-based duration model was developed to examine the effect of different driving volatility and traffic measures (i.e., relative speed, velocity standard deviation, and merging location) on merging GT, representing the aggressiveness of the merging events. Similar to the CF and crossing conflicts, merging crash risk was estimated using the EVT approach. The hazard model and the EVT model crash risk analysis revealed similar merging crash risks irrespective of the presence of AVs in merging events.

The findings of this Ph.D. dissertation identified critical areas for improvement that can be investigated further to enhance the safety performance of AV technology. The driving behavior analysis showed that current AVs need to adopt a more human-like driving style for future large-scale deployment. Standardized vehicle-to-vehicle (V2V) communication can warn conflicting HDVs equipped with low-level autonomous driving systems (ADS) technology and connected vehicle (CV) features about potential safety hazards to avoid unsafe interactions with AVs.

Persons with disabilities have the right to request and receive reasonable accommodation. Please call the Department of Civil and Environmental Engineering at 517-355-5107 at least one day prior to the seminar; requests received after this date will be met when possible.