Research

Traffic-State Reconstruction and Prediction under Partial Observability

MSC Lab, UGA  ·  Aug. 2025 – Present

A hybrid framework that couples deep learning with traffic flow modeling to reconstruct and forecast traffic states from partial observations.

• AI models that reconstruct short-term traffic occupancy and flow in urban signalized environments
• Traffic flow dynamics extend prediction to corridor-level horizons and capture traffic evolution beyond observed data

3D Digital Twin for Closed-Loop Simulation and Control of Autonomous Vehicles

MSC Lab, UGA  ·  Aug. 2025 – Present  ·  Sponsored by DOE, through ORNL

A multi-resolution digital twin integrating SUMO and CARLA for closed-loop simulation of traffic, control, vehicle dynamics, and hardware-in-the-loop.

• A five-layer (L0–L4) fidelity framework that quantifies the sim-to-real gap across geometry, sensor signals, perception, and behavior, with gap-budget attribution to trace error sources
• A ROS-free annotation toolkit for real-world multi-sensor logs (six cameras, 3D LiDAR) that auto-seeds 3D bounding boxes, reducing manual labeling to box refinement

Automated Vehicle Trajectory Modeling (IEEE ITSS AV Challenge)

MSC Lab, UGA  ·  Jun. 2026 – Sep. 2026

1st Place among 28 teams

Behavioral modeling of a following automated vehicle — predicting its 20-second future trajectory from the lead vehicle's trajectory and 10 seconds of its own history, evaluated on accuracy, safety, and comfort.

CDASim Upgrade for Cooperative Driving Automation Research

MSC Lab, UGA  ·  Sep. 2025 – Feb. 2026  ·  Sponsored by DOT, through Leidos

Core components of a distributed CARLA–CARMA–SUMO–ROS2 co-simulation framework unifying traffic, vehicle, sensing, and control stacks.

• Component- and system-level testing for scenario-based validation of cooperative driving automation (CDA)