Sample dataset · for partner review

Ego-centric bimanual manipulation data, ready for VLA training.

Explore manipulation skills → Browse scene coverage
Why this data

Built for the demands of embodied foundation models.

◆ Flagship · annotated
Annotated data

Our highest-quality labels: reconstructed 3D hand motion.

Manipulation skills

Dexterous, two-hand tasks — each with a full annotation spec.

Scene coverage

Bimanual trials across 30 real-world scenes.

Data specification

What ships with every clip.