JD open-sources EgoLive: 1,680-hour egocentric dataset for humanoid robots

CyberRobooo · x · 2026-08-14

JD has open-sourced EgoLive, a large-scale egocentric dataset recording real-world human tasks, aimed at humanoid robot learning.

It contains 1,680 hours of high-fidelity binocular video at 60fps, 65,866 data segments, and 346 distinct tasks, with data drawn from real business scenarios across retail, logistics, healthcare, and more. The post links to the project website and tech report.

Related event: JD Open-Sources EgoLive: 1,680 Hours of First-Person Video for Embodied AI(2 posts)→

Original post →

More from Embodied

Embodied channel →