HubEgo-100K v1.0 releases 10,124 hours of skilled-work egocentric video
HubEgo-100K v1.0 provides 10,124 hours of 1080p egocentric video captured by 12,227 contributors in 15,548 environments, with synchronized 100 Hz IMU data. The release contains 72,754 episodes with machine-generated action and language annotations, distributed in LeRobot and MCAP formats under CC BY 4.0. The arXiv paper says the dataset spans 11 setting types, 30 work categories, 56 task types and 35 action families, with 83% of footage recorded in workplaces. It does not yet include benchmark results; later versions are expected to add benchmarks and further releases toward 100,000 hours.
The paper presents the dataset as a scalable, openly licensed source of skilled-work video for robot learning, addressing the high cost of collecting robot data across diverse environments. Its diversity-focused design is intended to support policies that generalize across environments and objects.
Computer Science > Robotics
arXiv:2611.21938 (cs)
[Submitted on 16 Nov 2026]
Title:HubEgo-100K: In-the-Wild Egocentric Video of Skilled Work for Robot Learning
Authors:Javad Amirian, Armin Kiani, Loïc Schaeffer, Tim Sprecher, Mohamed Taha, Karthik Yadati
View a PDF of the paper titled HubEgo-100K: In-the-Wild Egocentric Video of Skilled Work for Robot Learning, by Javad Amirian and 5 other authors