Physics-Enhanced AI for Transportation

Building transportation AI that can be trusted.

We combine physical knowledge with data-driven learning so that automated vehicles and transportation systems stay safe and reliable in rare, unseen conditions — and we test them on real roads.

Vehicles, pedestrians and robots interacting with rainy, snowy and nighttime transportation environments.

Research Directions

VLM-MPC autonomous driving framework

Physics-Enhanced AI

Developing interpretable, physically consistent, and reliable AI methods by integrating physical domain knowledge with data-driven learning.

PINNVLM
CATS-V2V cooperative perception dataset

Automated Vehicles

Addressing safety challenges in complex and rare driving scenarios, with an emphasis on rigorous evaluation and real-world validation.

Field TestSafety EvaluationVehicle-in-the-Loop (ViL)
Park Street aerial vehicle trajectory dataset

Intelligent Transportation Systems

Developing intelligent solutions for transportation monitoring, infrastructure inspection, safety analysis, and traffic management to improve safety and efficiency.

Digital TwinTelematics Data

Selected Demonstrations

Robotic Manipulation

Mobile robotics and physical interaction

Autonomous Driving

Real-world eco-driving experiments

Human–Vehicle Interaction

Interactive digital twin environment

News

Our paper “A Low-Rank Method for Vision Language Model Hallucination Mitigation in Autonomous Driving” was accepted by Transportation Research Part C.

We are organizing the inaugural MIRAI Driving Workshop at ACCV 2026 (Osaka, Japan, Dec 15), with keynote speakers from Purdue University, Waymo, KAIST, and Hokkaido University.

Our paper “Physics-Enhanced Residual Policy Learning (PERPL) for safe cruising in mixed-traffic platooning” was published in Transportation Research Part C.

Received the WISL Award for Communicating Postdoctoral Research to the Public. UW Engineering feature.

Moments