Build AI

founder of @buildpbc

Egocentric-1M is an open dataset of roughly one million hours of first-person video captured from factory workers wearing Build AI's custom head-mounted glasses, published on Hugging Face under an Apache 2.0 license for free commercial use. It is aimed at robotics teams and foundation model researchers training manipulation policies, visuomotor models, and embodied agents that need dense human hand and eye data from real work, rather than staged lab demonstrations or third-person scene footage. The launch matters because it caps a rapid scaling arc from the same team. Build AI released Egocentric-10K in November 2025, Egocentric-100K in December, and now Egocentric-1M, which it describes as the largest egocentric dataset ever released, dwarfing every prior dataset combined . The earlier 100K release covered over 100,000 hours from 14,228 factory workers across Southeast Asia and 10.8 billion frames, captured in real production environments including assembly lines, sorting, packaging, and machining , and the 1M tier extends that approach to a scale that targets pretraining of general-purpose robot foundation models. Build AI is led by founder and CEO Eddy Xu. By April 2025, Xu had dropped out of Columbia to launch Build AI, which defines its mission as collecting human data to accelerate the deployment of general-purpose robots , and the company runs collection operations across SF, Shenzhen, and Bangalore. Framed by Xu as a step toward "the internet for physical AI," this release puts a corpus previously locked inside large labs into the hands of any team building robots that learn from how people actually do work.

Tags

Comments (12)