Research Projects
Generative Intelligence for Transportation Modeling
We develop deep generative models — including diffusion models, normalizing flows, GANs, and probabilistic mixture models — to address fundamental challenges in transportation data modeling. Our work spans traffic state estimation from sparse sensor observations, occupancy density estimation, trajectory generation, and probabilistic traffic forecasting. By leveraging generative approaches, we move beyond conventional deterministic methods to capture the inherent uncertainty and complex spatiotemporal correlations in transportation data. Related works
Vision-Language-Action Models for Autonomous Driving
We investigate Vision-Language-Action (VLA) models that integrate visual perception, natural language understanding, and vehicle control within a unified framework for autonomous driving. Our research explores how multimodal foundation models can interpret complex traffic scenes, reason about driving scenarios, and generate control actions. We also develop vision-language approaches for transportation surveillance tasks such as vehicle identification and traffic monitoring in real-world conditions. Related works
TMC-Agent: LLM-Powered Traffic Management
We are building AI agents powered by large language models (LLMs) to assist Traffic Management Center (TMC) operations. TMC-Agent aims to augment human operators by automating the interpretation of traffic network data, enabling natural language interaction with traffic management systems, and supporting real-time decision-making. Our research includes developing benchmarks to evaluate LLM capabilities on traffic network files and exploring how foundation models can be applied to transportation operations and planning tasks. Related works