Why this is hard
A model can read every sentence ever published and watch every video ever uploaded, and still not know how much force a drawer needs. Physical intelligence is not written down anywhere; it is performed, by people, and then it is gone.
Language models
Already collected, already public. The corpus existed before the models did.
Vision models
Images and video scraped from an internet that was already there. Again, found rather than made.
Physical AI
Nothing equivalent exists. Egocentric capture of ordinary human work has to be recorded on purpose, by someone, with consent.
The first two were found. The third has to be made.
Proof
Each Capture Partner's consent is written to an append-only ledger, where every entry is chained to the one before it by a SHA-256 hash, so an earlier entry cannot be changed without breaking every hash after it. A receipt proves that consent exists without revealing who the person is.
The record below is an example, because a real one names a real person. Its digest is produced by the same function that runs in production.
The eleven fields, in the order they are joined
The pre-image: those eleven, joined by |
loading…SHA-256 of that string
loading…The button below hashes the string above in your browser, with the Web Crypto API, and compares the result with the digest we published.
How it works
Five steps, shown on a real clip we publish in full. Each one links to the page that explains it, and the whole method is on How it works.
If the detector cannot run The recording stops at step 2, so no unblurred copy moves on.
—01 · Capture
For a commissioned brief, the task, the setting and the camera rig are fixed before anything is recorded. Drag the handle to compare the raw frame with the annotated one.
Consent receipt: none — recorded before the consent flow
02 · Redact
Faces the detector finds are blurred on our servers as the recording arrives. If the detector cannot run, the recording stops there, so no unblurred footage moves on.
03 · Segment
derived
Each clip carries an action code, so a dataset can be filtered by what happens in it. On this reference clip the cut is a fixed 120-frame window, shown at its true frame positions.
04 · Annotate
derived
21 joints per hand, each hand with a confidence score, on every tenth frame, so you can filter on quality rather than trust it. Drag the timeline to step through the sampled frames.
05 · Deliver
Delivered as LeRobot v2.1, HDF5 or FH-Ego v1 JSON, with Zarr v2 and TFRecord on request. Rehash the annotation file you download and you get the digest shown here.
Catalogue
Each brief below is a capture specification you can commission. Every one is delivered in one annotation schema, FH-Ego v1, so a dataloader you write once works across every delivery.
Price
Three things move the number, and each is fixed in the brief.
What moves the number · 1
An Indian household kitchen and an active distribution centre are not the same shoot. Access, scheduling and how many people have to consent all change with the room.
What moves the number · 2
How long a session runs, how many of them, and how many Capture Partners the brief needs to reach the volume you are asking for.
What moves the number · 3
Tier 1 is session metadata. Tier 3 is 21 3D joints per hand with per-hand confidence, and it costs more to produce and more to verify. You pick the tier in the brief.
Recording in India keeps a bespoke brief affordable.
Looking for paid recording work? This form is for labs buying data. Register as a Capture Partner →
For robotics and AI labs
Describe the policy and the embodiment you are training it for. We scope the capture specification, price it against your volume and format, and come back with what we can commit to, including the parts we cannot. Our team reads every one of these.
What a brief costs is scoped before it is priced: how we price, and what moves the number.
FAQ