Research focuses on transfer and generalization across domains and modalities for reliable perception: 3D detection/segmentation, zero-shot recognition, and robustness to distribution shift and adversarial perturbations across autonomous systems, vision-language, and video.