Research projects and systems I have developed or contributed to.
An end-to-end differentiable framework that learns ethical decision-making weights from naturalistic driving data. Validated on Waymo, NGSIM, HighD, and inD datasets. Designed for regulatory alignment with UNESCO and EU AI ethics frameworks.
A physics-constrained deep learning model that predicts vehicle trajectories by first inferring driver goals and then generating dynamically feasible paths via neural ODEs.
Fine-tuning video vision-language models (Qwen2-VL, VideoLLaMA2) for automated traffic crash analysis using a custom VQA dataset built from accident footage with curriculum learning.
Research within the Connected and Automated Vehicle Highway (CAVH) framework, including IRIS, VIU, and RCS subsystems. V2X communication data pipeline development for connected vehicle applications.
Fine-tuned Llama 3.1-8B model for detecting personally identifiable information in Wisconsin crash report narratives, enabling privacy-preserving data sharing for transportation research.
