Egocentric data

Egocentric data from real work.

Egocentric data is first-person video of hands-on work, recorded at real work sites.

Clips are captured on a smartphone, action camera or AR/VR headset, must meet the Operator Intent standard, and are scored for quality before they ship.

Egocentric
Egocentric capture: packing boxes
clips
Egocentric clips
hours
Hours recorded
3
Capture device types

Specification

What every clip contains.

Egocentric means the camera moves with the person, not a fixed room camera. The work is seen the way the person doing it sees it.
Capture devicesSmartphone, Action camera, AR/VR headset
FramingFirst-person, from the point of view of the person doing the work
SettingsReal work sites and everyday tasks, not staged sets
Example tasksPacking boxes (see the sample); full egocentric task list TBD
Resolution and frame rateTBD
Sensors beyond videoTBD
DeliveryDiscover: Browser download · Researcher: Browser download + API access · Enterprise: Browser + API + JSON download
AnnotationsTask prompt and validator quality score on every clip; further labels TBD

How we collect

Real places, one standard.

Work sites come with the lighting, clutter and object variety a model meets after deployment. Recording there puts that variety into the data; a single standard keeps every clip usable.
  • Recorded at real work sites, with real lighting, clutter and objects
  • Captured on a smartphone, action camera or AR/VR headset
  • Clips are held to the Operator Intent standard
  • Contributor capture through VLA Foundry is opening in waves via a waitlist
  • Custom egocentric collection is scoped with sales (Enterprise)

Quality and delivery

Each clip must meet the Operator Intent standard and is scored for quality before it ships. Clips are delivered by browser download, API or JSON export, depending on your plan.

How every episode is scored

FAQ

Frequently asked questions

What is egocentric data?

Egocentric data is first-person video of hands-on work, recorded from the point of view of the person doing it.

What is an egocentric video dataset used for?

Teams use egocentric video to teach models how people handle objects: grasps, tool use and the order of steps in a task. It is used in world model and VLA training, often alongside teleoperation data, which adds the robot's own actions.

Where and how is PrismaX egocentric data recorded?

At real work sites and in everyday settings rather than on staged sets, on a smartphone, action camera or AR/VR headset.

How is egocentric data quality checked?

Each clip must meet the Operator Intent standard, and clips are scored for quality before they ship. The full capture rules live in the PrismaX app.

Can I order egocentric data for specific tasks?

Yes. Custom egocentric collection is scoped with sales as part of Enterprise.

Next step

Egocentric data for your model.

Tell us the tasks and settings you need. Want to record clips yourself? VLA Foundry is opening egocentric capture in waves.