Catalog datasets

Off-the-shelf datasets. Sample today, license by the hour.

Rights-cleared demonstration sets, organized by activity domain. Pull a free evaluation slice, test it against your loader, then license only the hours you need — priced by tier.

How it works

Browse, sample, license.

Browse by domain

Pick an activity domain — cooking, assembly, cleaning, retail handling, office manipulation. Each set is egocentric, real-world, and consented at source.

Pull a free sample

Get a free evaluation slice within two business days. Load it, run it against your task spec, and check tier and format before committing.

License by the hour, by tier

License the hours you need at the tier you need — Raw + filtered, Structured, or VLA-ready. No scoping call, delivery in days.

Domains in the first release

The initial 5,000-hour catalog spans five everyday-manipulation domains. Per-domain hour counts are being finalized with our first design partners.

Domain What it covers Hours Status
Cooking & food prep Chopping, stirring, plating, clearing a work surface. TBD Available
Assembly & tool use Screwdriving, fastening, assembling flat-pack and small parts. TBD Available
Cleaning & household Wiping, folding, sorting and tidying household spaces. TBD Available
Retail & handling Picking, packing, shelving and order handling. TBD Available
Office & manipulation Fine manipulation of small objects at a desk. TBD Available

Per-domain hours are being finalized. Request a sample to see current availability for your domain.

What each set includes

Every catalog set ships with the base layers; structure above that depends on the tier you license.

  • ✓Egocentric video, deduped and quality-filtered
  • ✓Sensor metadata sidecar (timestamps, IMU)
  • ✓Clip segmentation into discrete, task-aligned clips
  • ✓Pose keypoints, contact events and narration (Structured tier and above)
  • ✓Action labels and LeRobot / RLDS packaging (VLA-ready tier)
  • ✓Consent record and provenance for every clip

Why catalog is the low-touch option

No scoping call, no written spec, no pilot batch. The domains already exist, already cleared QC, and already carry commercial training rights. You sample, you license, you load — usually in days.

When to commission instead

If you need specific tasks, environments or distributions that aren't in the catalog, custom collection runs from 100 to 5,000 hours against your written spec.

Need demonstration data?

Book a 30-minute data review. We'll scope your task taxonomy and send a free sample slice.

Request a free sample slice