Gen-HumanEgo: Open-Source Egocentric Human Data for Robot Learning

Loading video

Loading video

GenRobot released Gen-HumanEgo, an open dataset of 1,800+ hours of first-person human demonstrations collected across 10,000+ different tasks with a unified six-camera setup. Scale alone is not the claim: GenRobot's Data Foundation Model (DFM) converts the synchronized multi-view ego video into training-ready signals, where video, action, 3D hand reconstruction, depth and semantic annotations are precisely aligned and encoded into unified tokens so downstream models train against one shared representation. The stated goal is turning real-world human experience into data that embodied AI models can actually learn from, with Gen-HumanEgo as the first release.