Our group investigates the complex interactions between humans and transportation systems, with a particular emphasis on the psychology, behavior, and safety of pedestrians, drivers, and other transport stakeholders. By integrating immersive virtual reality (VR) simulation, advanced data mining, and state-of-the-art artificial intelligence (AI) modeling, we examine how individuals perceive, decide, and act in diverse and safety-critical mobility contexts. Our research spans behavioral analysis, large-scale accident data modeling, and AI-driven traffic system prediction, enabling us to translate empirical insights into evidence-based strategies and decision-support tools for safer, more adaptive, and human-centered transport systems.
Our interdisciplinary approach draws on transportation engineering, behavioral psychology, data science, and machine learning, bridging fundamental human factors research with advanced computational modeling to address real-world transport safety challenges.
We explore how cognitive, perceptual, and cultural factors shape pedestrian and driver behaviors, particularly in safety-critical situations.
Pedestrian behavior under unfamiliar traffic rules – Using immersive VR simulations, we have identified habitual looking behaviors that increase pedestrian risk when crossing in traffic systems opposite to their home country’s driving rules.
Alcohol impairment and decision-making – Our VR experiments demonstrate how intoxication shifts pedestrian crossing motivations from safety to efficiency, increasing collision risk.
We develop statistical and computational frameworks to uncover hidden patterns in transport safety data, particularly in under-researched contexts.
Bayesian random-parameter modeling – We proposed a Bayesian random-parameter generalized beta of the second kind model with heterogeneity in means to accurately capture the skewed, heavy-tailed nature of economic losses from fishing vessel accidents.
Quantile regression for heterogeneous effects – Our Bayesian random-parameter quantile regression with heterogeneity in means (BRPQRHM) framework reveals how vessel, environmental, and accident-related factors vary across loss quantiles, informing targeted maritime safety policies.
Text mining of safety reports – Using structural topic modeling, we analyze over 12,000 police narratives of non-collision bus injuries to identify recurrent causal patterns and propose tailored countermeasures.
We advance machine learning and physics-informed methods to model, predict, and optimize transportation systems.
Inverse reinforcement learning (IRL) – We apply IRL to quantify the latent safety–efficiency trade-offs in pedestrian crossing behaviors, enabling high-fidelity behavior modeling under different risk conditions.
Physics-Informed Neural Networks (PINNs) – Our distance-informed neural Eikonal solver addresses the limitations of discretization in solving dynamic traffic assignment PDEs, achieving higher accuracy in heterogeneous cost fields.
Risk-aware pedestrian–vehicle interaction modeling – We proposed a Bayesian-optimized Social Force Model with risk perception and adaptive decision-making, improving both trajectory accuracy and behavioral realism across diverse traffic scenarios.
By combining behavioral science, accident analytics, and AI-driven modeling, our goal is to transform empirical behavioral insights into predictive and prescriptive tools for transport safety. We aim to support policy-makers, urban planners, and industry stakeholders in creating traffic environments that are safer, more adaptive, and human-centered.