Meet the PrismaX team at CoRL in Austin, November 9. RSVP →

Solutions · World model data

World model data

World model data is video of real physical interaction — hands and robots manipulating objects, seen from the actor’s point of view and from fixed cameras — from which a model learns to predict how a scene changes in response to actions.

PrismaX supplies it as egocentric video from real work sites and as multi-view teleoperation episodes with synchronized actions.

Talk to salesBrowse the data

The data problem

What gets in the way

Video without actions

Plain video shows what changed but not the action that caused it. Models can infer latent actions, but recorded actions remove the guesswork.

The wrong viewpoint

Room-scale footage misses what the hands are doing. Interaction is clearest from cameras on the head and wrists.

Artifacts that get learned

Dim light, noise and compression artifacts become features the model learns to look for instead of the scene itself.

Repetitive scenes

The same few tasks in the same few rooms don't teach a model how the wider physical world behaves.

What the data needs

Interaction, from where it happens.

  • Continuous episodes of hands-on physical interaction, from start to finish of a real task.
  • Egocentric viewpoints — head and wrist cameras — alongside a fixed overhead view of the workspace.
  • Synchronized actions or joint states wherever the embodiment is a robot.
  • Many scenes and objects: homes, kitchens, warehouses, factories, labs and offices.
  • Clean, evenly lit images without sensor noise or heavy compression.

How PrismaX provides it

From the catalog, or collected to spec.

Robotics Data

Egocentric data

First-person video of people doing hands-on work at real work sites, recorded on phones, action cameras and headsets.
Learn more Robotics Data

Teleoperation data

Multi-view robot episodes — overhead and wrist cameras — with every joint angle recorded.
Learn more Services

Custom data collection

Scenes, tasks and volumes collected to your spec. Egocentric hours available per month TBD.
Learn more
We work with teams developing world models, vision-language-action models and other physical AI systems, as described in Introducing PrismaX Robotics Data. For how video-trained models learn actions without labels, see GR00T Part 2: Labels Without Labels.

Industries and scenes

Where the demonstrations come from

Catalog scenes that apply to this use case. Other environments are collected to spec.
  • HomeTidying, sorting, folding, placing
  • KitchenFood handling and clean-up
  • LogisticsPacking, unpacking, bin picking
  • IndustrialComponent handling
  • OfficeDesk and supply handling
  • Lab and ResearchBench work and lab procedures

Related tasks

  • Biology LabLab and Research · 108.8k
  • Pick and place various fruits into containersKitchen · 2.8k
  • Pick and place flowers into vase and vase on shelfHome · 2.0k
  • Block manipulationHome · 2.0k
  • Pack box of componentsLogistics · 1.8k
  • Pick and place slippers into binHome · 1.6k
  • Tying shoelacesHome · 1.5k
  • Clean up kitchen areaKitchen · 1.4k
Browse the full catalog

Sample videos

See the episodes

All sample videos
Egocentric
Egocentric capture: packing boxes

First-person clips of hands packing boxes, recorded from the point of view of the person doing the work.

Teleoperation
Bimanual flower arranging, overhead and wrist cameras

Two robot arms arrange flowers in a vase, seen from the overhead and wrist cameras at the same time.

Related reading

All news

Perspective

Not All Robotics Data Is Created Equal

A primer on the two paths of physical AI, the data each one actually needs, and what data quality looks like at the technical level.

Perspective

Standards for Data

The first arc of the PrismaX Standards for Data: repeatable guidelines for collecting pretraining-grade robotics data, and the reference our validators use to apply them.

Deep Dive

Intro to AI for Robotics

An introduction to AI-native robotics, covering learning loops, teleoperation data, and how modern robots move from demos to real work.

FAQ

Frequently asked questions

What is world model data?

Video of real physical interaction from which a model learns to predict how a scene changes in response to actions: egocentric video of people working with their hands, and robot episodes with synchronized camera views and joint states.

Why egocentric video for world models?

Cameras on the head and wrists see the interaction itself — the grasp, the contact, the object moving — which room-scale footage often misses. It is also cheap to capture at scale and the motion is smooth.

Do your videos include actions?

Teleoperation episodes record the angle of every joint alongside synchronized overhead and wrist video.

How is quality controlled?

Every episode passes automated checks and then human validators review it against our Standards for Data, including full view of the workspace and clean, evenly lit images.

Next step

Show your model the real world.

Tell us the scenes and the volume. We'll match it to the catalog or scope an egocentric collection.

Talk to salesSee pricing