EgoSuite-Open100K

EgoSuite-Open100K is a staged 100,000-hour open release of egocentric human data from Lightwheel, positioned as the largest fully-annotated open egocentric human dataset for Physical AI. Released in partnership with Hugging Face in August 2026, the dataset is published in stages: the full 100,000 hours are not yet downloadable at once, with the first public batches coming online progressively. At full scale it spans 100,000 hours across 15,000+ tasks and 15,000+ scenes. Tasks span 18 task categories and 7 real-world environment categories: Home (40%), Hospitality (10%), Retail (10%), Sports (10%), Logistics (10%), Office (10%), and Industry (10%). Example tasks include packing, sorting, wiping, folding, and pick-and-place operations across household, retail, and industrial settings. The data carries hand pose, body pose, and event-level semantic annotation, with wrist camera coverage on part of the set, and ships in LeRobot v3 and MCAP formats. It is organized as a Hugging Face collection with three lines: EgoStandard (90,000-hour head-view line), EgoPro (10,000-hour head-and-wrist line), and EgoDemo (a 50-hour developer sample covering every annotated subset plus two raw-video variants). Access on Hugging Face is gated (manual approval) under a commercial-training-no-resale license: open for academic research and commercial training alike, with part of the data requiring an access application. EgoSuite-Open100K is the Behavior layer of Lightwheel's Physical AI data engine, complementing its SimReady world assets and RoboFinals evaluation platform. As an open egocentric human demonstration dataset targeting robot learning, it is a reference node for embodied AI dataset and human demonstration data collections.