Hub open-sources HubEgo-100K, a 10,000 hour first-person work dataset
Hub has open-sourced HubEgo-100K, a first-person, head-mounted video dataset of skilled manual work, released in stages toward 100,000 hours.
The first release has 10,124 hours, 72,754 episodes, and 12,227 contributors, filmed in real workplaces and homes, not staged lab sessions. It covers 62,000+ task descriptions across 56 task types, with 1920x1080 video at 30 fps, a 100 Hz IMU, and per-episode task, step, difficulty, and environment notes, in LeRobot v3 and MCAP formats.
It is free under CC BY 4.0, including commercial use with credit. Access requires a short form on the Hugging Face dataset card.