The sensing and data layer for Physical AI
Foundation models for robots are starving for data. The internet has trillions of words but almost no first-person footage of someone folding laundry or using a drill the way people actually do.
So we put a field team on the ground and fit participating workers with compact rigs. They repair, assemble, make and move things like they do every day, and the cameras catch the work from their point of view.
You can think of us as "staffing agency" meets "data pipeline".
We take on anything from a single scripted scenario to longer capture programs across multiple sites. If a human can do it, we can film it.
Auto Service
Garment Work
Home Appliance Repair
Phone Repair
Send us your data spec. Which scenarios, which environments, how many hours. We can start with a focused pilot capture.
We coordinate the workers, the sites and the compact rigs. Clear briefs, repeatable setups.
Workers record through their normal shifts and every session goes through QC. If hands or the task leave frame, unusable footage is flagged.
Footage ships reviewed and organized in straightforward batches, with clear filenames and capture notes.
AI data platforms that need a reliable field-collection partner for embodied AI projects.
Humanoid labs training manipulation policies who need more human demonstrations than their own teleop rigs produce.
Egocentric AI teams building smart glasses and world models who need thousands of hours of first-person life.
First-person footage of skilled work in real operating environments, with hands, tools and task progress kept central in the frame.
Bring us a task list, target environment and hour count. We coordinate sites and workers, record to the brief and review footage before delivery.
Footage organized by task and session, ready for your team to inspect, annotate or process inside its own data pipeline.
MP4/MKV video, MCAP, HDF5, RLDS/TFDS, LeRobot-compatible datasets, or raw synced streams with calibration files. Tell us your pipeline and we'll match it.
Cameras on simple head- or chest-mounted rigs, configured around the task and site. Custom camera placement is available where the work requires it.
Yes. Every worker signs a release covering commercial ML training use, and bystanders are handled by location agreements or anonymization. Full provenance documentation ships with every dataset.
Faces, screens, license plates, and bystander audio are blurred or redacted before delivery by default. We can also deliver raw-with-releases where your use case requires it.
Yes. Send the spec and we'll scope the task, site, viewpoint and volume, then start with a focused pilot capture.
You choose the license: exclusive buyout, non-exclusive license, or subscription to our growing off-the-shelf library. Exclusivity is priced accordingly.
Per hour of delivered footage, priced by sensor loadout and exclusivity. Samples are free and come with an exact quote.
Tell us what you need and we'll scope a focused pilot capture.
Get a Sample