Multimodal data lab and global contributor network that captures, annotates, and delivers real-world training data for AI and robotics.
Hub is a multimodal data lab that captures, enriches, annotates, and QA's real-world data for frontier AI labs and top robotics companies through a global network of 150,000+ contributors and 730+ commercial sites across 150 countries. Data types include RGB-D egocentric video with 6-DoF VIO and 3D hand pose, object and hand tracks, temporal action segmentation, dense scene captions, and multilingual audio with speaker diarization and emotion labels. Bespoke data collection can start within hours through automated contributor matching and diversity enforcement. Hub is a product of Hub.