Initiated and developed at Inchor (SaferDrive AI). LLM/VLM agents mutate real logged drives into safety-critical corner cases, replay them in closed loop, and a video world model renders the result.
A dynamic robotaxi routing framework that turns centrally controlled fleets into cooperative drive-by sensors for urban traffic monitoring. Under revision at Transportation Research Part C.
Project leader of a two-phase cooperative perception project: a V2X cooperative perception safety study funded by NSF (Transportation Research Record, 2024), then an infrastructure-vehicle cooperative perception system funded by CCAT.
Behavior-level reconstruction of human driving, from physics-constrained prediction to cognitively interpretable response: IDM-Follower (IEEE Transactions on Intelligent Vehicles, first author) and Neural Force (manuscript in preparation).
MMITSS, an open-source connected-vehicle signal priority system: deployments at Mcity and on Ann Arbor’s Ellsworth Road corridor (USDOT FHWA), MMITSS for the Detroit QLine (WSP for Michigan DOT), integration into FHWA’s CDASim co-simulation (Leidos for USDOT FHWA), and a signal-aware connected vehicle.
Core developer of TeraSim, an open-source city-scale traffic simulation platform for AV safety testing: MotionTransformer-based naturalistic driving behavior (NNDE), and a city-scale NVIDIA Metropolis Blueprint demo with NVIDIA and the City of San Jose that was selected for the GTC 2026 keynote.
Independent research at Inchor (SaferDrive AI): a reinforcement learning waypoint controller for a Boston Dynamics Spot robot in NVIDIA Isaac Sim, with a Control Barrier Function safety filter that keeps a dynamic safety boundary around moving obstacles.