University of Minnesota · Choi Lab
AI that turns messy city data into operational insight.
We're the Choi Lab in Civil, Environmental, and Geo-Engineering at the University of Minnesota, Twin Cities — building AI for urban mobility with deep generative models, vision-language-action models, and LLM agents, in service of more sustainable, efficient transportation systems.
Research areas
All projects →Generative Intelligence for Transportation Modeling
Diffusion, normalizing flows, GANs, and probabilistic mixtures for traffic state estimation, trajectory generation, and probabilistic forecasting — beyond deterministic point estimates.
AI-Powered Connected and Automated Driving
Multimodal foundation and Vision-Language-Action models, plus deep RL for cooperative driving and vehicle control — extending to traffic-surveillance tasks like vehicle identification and monitoring.
TMC-Agent · LLM-Powered Traffic Management
LLM-powered agents that augment Traffic Management Center operators — interpreting network data, enabling natural-language interaction, and supporting real-time decisions.
Recent publications
All publications →
Toward Safe Integration of UAM in Terminal Airspace: Route Feasibility via Probabilistic Aircraft Trajectory Prediction
PMA-Diffusion: A Physics-guided Mask-aware Diffusion Framework for Traffic State Estimation from Sparse Observations
BO4Mob: Bayesian Optimization Benchmarks for High-Dimensional Urban Mobility
- A Gentle Introduction & Tutorial on Deep Generative ModelsTransportation Research Part C · 2025
- NextSim: Multi-Level Traffic Simulation for Urban NetworksIEEE Trans. ITS · 2026
- Scalable Dynamic Mixture Model for Probabilistic ForecastingTransportation Science · 2025
- A Survey on Vision-Language-Action Models for Autonomous DrivingICCV Workshops · 2025
Highlighted projects
All repos & data →TrajGAIL
Urban vehicle trajectory generation via generative adversarial imitation learning (TR-C 2021).
Benchmark · ODBO4Mob
Bayesian optimization benchmarks for high-dimensional urban mobility (NeurIPS 2025 D&B).
SurveyDGMinTransportation
Companion code & notebooks for the TR-C 2025 deep generative models survey.
GenAITrajFlow
Normalizing-flow framework for occupancy density estimation from trajectories.
LLMTrafficNetQA
QA benchmark for evaluating LLM performance on traffic network files (SIGSPATIAL 2025).
Vision · Sensingcamera2detector
Real-time camera validation of MnDOT highway detectors using YOLO computer vision.
VLA · SafetyHighwayVLM
Open-source highway-safety analysis with vision-language models.
ODODS_PLD
Analytical OD estimation via NNLS and Projected Langevin Dynamics, validated under SUMO.
LLMSynPopPred
LLM-based synthetic population generation — IPF, CTGAN, DistilGPT-2, and Llama 3.1 8B.
The team
Full team & alumni →
Seongjin Choi
Guoliang Feng
