Enriched egocentric video
First-person task footage with depth, pose, segmentation and action-language pairs.
Rights-cleared, quality-scored training data and real-world evaluation for physical AI.
Robots can't learn from the internet. Every example has to be made in the physical world, by a person, on camera, with consent. That's what we make.
What we deliver
First-person task footage with depth, pose, segmentation and action-language pairs.
Bimanual robot episodes with force and tactile streams, including deliberate failure and recovery.
Repeatable benchmark campaigns for sim-trained models, with a failure taxonomy you can act on.
How we work
Plain-language consent in local languages, a registry per clip, faces blurred by default.
Collectors paid above market, told what the data trains, free to opt out.
A sample in your format, with data dictionary and QA report, before you sign anything.