Publications
Work on autonomous driving, motion forecasting, and computational imaging.
Multi-dataset training and BEV-Forcing for zero-shot transfer of driving vision-language-action models across camera rigs, studying how spatial supervision and training diversity affect generalization.
A structure-aware approach to reasoning about future interactions in multi-agent trajectory prediction.
Preference optimization for more coherent predictions across interacting agents in complex traffic scenes.
Deep image priors for reconstructing three-dimensional volumes from stacks of differently focused microscopic images.
Using road-map structure as a self-supervised signal for learning representations for motion forecasting.