|
University of Maryland (UMD) civil and environmental engineering doctoral candidate Kaitai Yang has been named the 2026 winner of the department’s Best Doctoral Research Award.
Yang won the award for his dissertation, A Hybrid Framework for Modeling Surrounding-Vehicle Behavior for Evaluating Autonomous Vehicle Performance in Risk-Critical Driving Scenarios. His advisor is Associate Professor and Clark Faculty Fellow Xianfeng (Terry) Yang.
His research aims to position autonomous-vehicle (AV) safety assessment as an interaction-centered traffic simulation problem, rather than an ego-vehicle performance problem alone. In risk-critical scenarios, Yang explained, surrounding human-driven vehicles are not simply background traffic; their responses to adverse weather, cut-in maneuvers, and leader-transition events shape the conflict dynamics and safety margins through which AV performance is evaluated.
By developing behaviorally interpretable and data-supported models of these responses, Yang’s work strengthens the microscopic behavioral foundation of safety-oriented simulation and supports more credible evaluation of future AV systems.
“I am honored to receive this recognition from the CEE department and grateful for the opportunity to share my doctoral research with the broader engineering community,” Yang said. “This work has been shaped by the guidance and support of many people, especially Professor Yang, my committee members, collaborators, and the CEE community. I am very thankful for their mentorship throughout my Ph.D. journey.”
Below is an abstract of the award-winning Ph.D. dissertation.
Kaitai Yang
A Hybrid Framework for Modeling Surrounding-Vehicle Behavior for Evaluating Autonomous Vehicle Performance in Risk-Critical Driving Scenarios
Effective autonomous-vehicle (AV) safety assessment requires realistic representation of the surrounding traffic environment, particularly in risk-critical scenarios involving adverse weather, abrupt cut-ins, lane-changing interactions, and close vehicle-following conditions. Traffic simulation provides an essential setting for evaluating such scenarios at scale, while its credibility depends on whether surrounding human-driven vehicles are modeled in a behaviorally plausible, interpretable, and data-supported manner. This dissertation addresses that challenge by developing a hybrid framework for modeling surrounding-vehicle behavior under complex and safety-critical driving conditions.
The research integrates microscopic traffic-flow theory, vehicle trajectory data, optimization-based calibration, Bayesian uncertainty quantification, and data-driven trajectory prediction. Mechanistic car-following models are extended to represent more conservative driver responses under rain and snow, as well as the gradual transition of driver attention during cut-in maneuvers. These models are calibrated and validated using observed vehicle trajectories to improve the representation of speed, spacing, and leader-following behavior. The framework further integrates data-driven trajectory prediction to represent richer future vehicle motion in freeway driving scenes while incorporating traffic-flow-relevant interaction information. Together, these contributions strengthen the behavioral foundation of traffic simulation for AV safety assessment and support more credible transportation safety analysis and future autonomous-vehicle evaluation research.
July 16, 2026
|
