Works across 2D and 3D computer vision and multimodal perception for segmentation, tracking, and real-time human-robot interaction, combined with uncertainty estimation and out-of-domain detection in deep learning to enable practical AI safety and dependable automation.