HubEgo-100K: In-the-Wild Egocentric Video of Skilled Work for Robot Learning
Summary
HubEgo-100K presents a large-scale in-the-wild egocentric dataset of skilled work for robot learning: 10,124 hours of 1080p video with synchronized IMU data from 12,227 contributors across 15,548 environments, released as 72,754 annotated episodes in LeRobot and MCAP formats to study diversity-driven policy generalization.
View Cached Full Text
Cached at: 11/19/26, 03:40 AM
Paper page - HubEgo-100K: In-the-Wild Egocentric Video of Skilled Work for Robot Learning
Source: https://huggingface.co/papers/2611.21938
Abstract
Robotpoliciestrainedbyimitationgeneralizepoorlyoutsidetheenvironmentsandobjectsseenintraining.Publisheddata-scalingstudiesfindthatgeneralizationgrowswiththediversityoftrainingenvironmentsandobjectsratherthanwithmoredemonstrationsperenvironment.Robotdataiscostlytocollectacrossmanyenvironments,whereasegocentrichumanvideocanberecordedinthewildatscale;existingegocentricvideo,however,mostlyshowsdaily-lifeactivityandcontainslittleskilledworkinrealworkplaces.WepresentHubEgo-100K,adatasetbuiltfordiversity.Thisversionofthepapercoversthefirstrelease,v1.0:10,124hoursof1080pegocentricvideowithsynchronized100Hzinertial(IMU)datafromonehead-mountedphonecamera.Inv1.0,12,227contributorsrecordtheirownunscriptedworkin15,548environments(eachonecontributor’sownsettingofonetype,suchastheirkitchenorworkshop)across11settingtypes,30workcategories,56tasktypesand35actionfamilies,with83%ofhoursinworkplacesettings.Thereleasecomprises72,754episodeswithmachine-generatedactionandlanguageannotations,inLeRobotandMCAPformatsunderCCBY4.0.Wemeasurediversityalongeachaxis,showthatinsidev1.0,withepisodesinrandomorder,newcontributorsandwordskeepappearing,andrelatethemeasurementstopublishedevidenceonsixdatapropertiestiedtopolicyperformance,includinghandvisibilityandinertialsensing.Wealsodocumentcollection,curation,privacymeasuresanddataformat,andprovideadatasheet.Laterversionsofthispaperwilladdbenchmarksandcoverfurtherreleasestoward100,000hours.
View arXiv pageView PDFProject pageAdd to collection
Get this paper in your agent:
hf papers read 2611\.21938
Don’t have the latest CLI?curl \-LsSf https://hf\.co/cli/install\.sh \| bash
Models citing this paper0
No model linking this paper
Cite arxiv.org/abs/2611.21938 in a model README.md to link it from this page.
Datasets citing this paper1
#### HubLabs/HubEgo-100K LeRobot• Updatedabout 2 hours ago • 72.8k episodes • 30.8k • 46
Spaces citing this paper0
No Space linking this paper
Cite arxiv.org/abs/2611.21938 in a Space README.md to link it from this page.
Collections including this paper1
Similar Articles
@LeRobotHF: 100,000 hours of human hands doing real work. LeRobot format, fully annotated and open. EgoSuite-Open100K from @Lightwh…
LightwheelAI and Hugging Face have released EgoSuite-Open100K, a large-scale open dataset of 100,000 hours of egocentric human activities, fully annotated and available in LeRobot format for AI research and training.
@AdinaYakup: Gen-HumanEgo New dataset for robot learning from @GenrobotAI - 1,800+ hours of egocentric human data - 10K+ unique task…
Gen-HumanEgo is a new dataset released by GenrobotAI for robot learning, featuring over 1,800 hours of egocentric human data with 10K+ unique tasks and synchronized camera views including 3D hand keypoints and structured annotations.
HumanNet: Scaling Human-centric Video Learning to One Million Hours
HumanNet is a large-scale human-centric video dataset with one million hours of annotated footage, designed to train vision-language-action models. It demonstrates that egocentric human video can effectively replace robot data for embodied intelligence tasks.
Ego2Robot: Scalable Robot Data Synthesis from Egocentric Human Data
Ego2Robot is a scalable pipeline that converts egocentric human manipulation videos into robot training data via action retargeting and visual synthesis, producing 18,561 hours of data across 15 robot morphologies. Experiments show that joint pretraining on this synthesized data improves out-of-distribution generalization for vision-language-action models, including on real-robot deployment.
HumanScale: Egocentric Human Video Can Outperform Real-Robot Data for Embodied Pretraining
This paper finds that egocentric human video, when processed with a filtering and labeling pipeline, can outperform teleoperated real-robot data for pretraining embodied foundation models, achieving lower validation loss and higher success rates on real-robot tasks.