Multi-view video · Hand tracking · Collected to order
Real-world data for physical AI.
Multi-view video and hand-tracking data from real production lines, for training robot policies and world models. In stock or collected to order, consented and de-identified.
Illustrative renders
2,500 hrs / week
QA-passed collection capacity. Only usable episodes are counted.
Penn·Columbia·NYU
Built by a team with deep roots in machine learning and robotics.
Consented
Every operator signs a release covering model training.
De-identified
Faces and PII redacted before data leaves the site.
Synchronized
Hardware-timestamped video, motion and telemetry.
Your format
LeRobot, RLDS, MCAP, HDF5 or WebDataset.
Catalog
Datasets.
Hours are accepted hours: only QA-passed footage is counted and billed.
In stock
Recorded and QA-passed. Licence now.
Collected to order
Recorded in the settings and tasks you specify.
Custom collection
Need something not listed?
Tell us the scenes, tasks and sensors. We pilot, then scale.
- 1
Specify
Scenes, tasks, sensors and volume.
- 2
Pilot
A small paid batch you validate in your own pipeline.
- 3
Scale
Full collection, QA'd and delivered with per-episode provenance.
Contact
Tell us what you’re training.
We reply with a datasheet, a sample clip and pricing. Or email info@kenwork.ai.