Foundation Dataset¶
46,893 real egocentric videos captured by the Humanoid Network community, representing more than 103 million frames.
The Foundation Dataset was captured by our community and is used to build the HanVerse world foundation model and Hydra validation stack. It gives Hydra the grounding it needs to place and score new robot motion intelligently.
What it shows¶
| View | What you see |
|---|---|
| Hydra neural map | A behavior-space view of the embedded source episodes. |
| Episode inspector | A selected episode with its task, preview, and metadata. |
| Metrics | Corpus size, embedding state, behavior families, and read-only status. |
Why it matters¶
Hydra needs a foundation before it can judge a brand-new motion. The Foundation Dataset is that foundation: 46,893 real community-captured egocentric videos, grouped into behavior families and embedded in a shared 3D behavior space, so the HanVerse world foundation model and Hydra can place a new motion in context instead of scoring it blind.
What it isn't¶
The videos are real community contributions, not synthetic placeholder content. The Dataset tab is a read-only view of that captured source corpus; it is separate from the app's current simulated Motions feed, and it is not a reward surface or public data-export portal.