Robots can learn from the human point of view.
First-person video shows what a person sees while they act. We supply it from contributors who have agreed to how it is used, licensed, quality-checked and delivered with a provenance record.
Learning from what people do.
Robots that learn from demonstration
Seeing tasks done by people, from the viewpoint a robot has.
World models based on real settings
Learning how objects, hands and environments behave in real settings.
Vision-language-action models
Pairing what is seen with what is said and done.
The view a robot actually has.
Egocentric video shows hands, objects and tasks from the actor's own point of view. Video can be paired with supporting data such as audio or sensor streams, depending on what a project needs and what contributors have agreed to.
Consent covers more than the person filming.
First-person video captures the wearer, the places they are in and sometimes other people. We treat all three.
| Contributors | Each contributor agrees in writing to how recordings are used and licensed before they record. |
|---|---|
| Other people in frame | Our aim is that delivered data does not identify bystanders. |
| Locations | Recordings happen only where the contributor has permission to record. Sensitive places are excluded. |
| Quality | Recordings are reviewed against the project specification before delivery. |
| What you receive | The data, license terms, and a provenance record linking each recording to its consent and processing history. |
Pair the data with experts and a platform.
Expert validation and evaluation
Domain experts can check the data and test the models trained on it.
Platform workflows
Annotation, review and governance records in one place, from data to deployment.
Scoping a robotics or world-model dataset?
Tell us the tasks, environments and modalities you need.