Papers marked * are first-author. See also: Google Scholar
Under Review / In Progress
Object-Trajectory Narratives: A Structured Motion Grammar for Language Understanding of 3D and Virtual Worlds*
Distributed 3D scene data pipeline from engine telemetry (no pixels), enabling 7B LLMs to narrate complex motion with GPT-4V-level accuracy at a fraction of the compute.
Temporal VOS Loss: Enforcing Smooth Object Motion for Video Object Segmentation
A new loss for video object segmentation enforcing smooth object motion, yielding up to 4% J&F gains on small, occluded, and reappearing objects across DAVIS, YT-VOS, and LVOS.
TeachPilot: A Classroom LLM Assistant with Instructor-Controlled Help Levels*
A classroom assistant with instructor-controlled prompts and help levels to promote think-first engagement; statistically significant gains in student AI literacy.
2025
2024
Wheeler: A Three-Wheeled Input Device for Usable, Efficient, and Versatile Non-Visual Interaction*
2023