Capture • Dataset • Evaluation
EgoVerse 
Human data from around the world, built for robot learning
EgoVerse is an ecosystem for curating, accessing, and learning from human data for robot learning. It hosts a "living" dataset, continuously expanded by the consortium, and driven by a research community advancing rigorous science of human-to-robot transfer across tasks and embodiments.
What EgoVerse Captures
Dense annotations and diverse task coverage
Ready for Policy Learning
Includes accurate camera poses, 3D head tracking, and dense language annotation.
Standardized and Open-Ended Tasks
EgoVerse contains standardized flagship tasks alongside open-ended scenarios across diverse tasks, scenes, objects, and operators.
Built for Human-to-Robot Transfer
Validated across diverse robot embodiments and tasks
Georgia Tech (RL2 Lab)
Task: object-in-container • ID/OOD
Stanford University (REAL Lab)
Task: cup-on-saucer • bimanual precision
UC San Diego (Wang Lab)
Task: bag-grocery • long-horizon
ETH Zurich (CVG & SRL Labs)
Task: bag-grocery • long-horizon
Consortium Partners
Built across academic and industrial partners
A Living Dataset
Data from around the world, continuously growing.
Dataset Snapshot
1,362
Hours of human demos
~80k
Episodes
1,965
Tasks
240
Scenes
2,087
Unique demonstrators
Stats from the current release.