Dataset catalog

Coverage, stated as cells filled — not as hours sold.

Hours are a vanity metric on their own. What determines whether a policy generalises is how many environment × condition cells you actually hold, so that is what we publish — including the cells that are still thin.
Validated hours
1,247 shipped
Environments
6 shipping, 4 thin
Enrichment
On every clip, no tiers
License
Commercial, buyer-owned

Hours by environment

What we hold today.

Validated hours only: footage that passed sync and QA and has shipped to at least one buyer. Recaptured and rejected hours are excluded rather than quietly counted.

Validated hours by scenarioas of 2026-07
01k2k3kKitchen & food prep2,840Household chores2,310Tool use & assembly1,680Warehouse & logistics1,420Retail & shelf ops890Automotive service640Construction & trades410Clinical & care180Agriculture120Hospitality95VALIDATED HOURS
DEEP — SHIPPING NOWTHIN — ACCEPTING SPEC

Condition cells

Where the gaps are.

Percentage of target cells filled per environment × condition. Anything under 40 percent is a gap we will tell you about before you scope a program against it.

Coverage matrix · environment × conditionindex 0–100
DaylightMixedLow lightClutteredMulti-personFailure cases94886179442890825571512272763866581968713362641477644183293158421247369KitchenLiving spaceWarehouseRetail floorWorkshopOutdoor site
Index = validated hours in that cell, normalised against our deepest cell. Failure cases are deliberate mistakes, drops, and recoveries — the hardest thing to source and the column we are shortest on.

Thin coverage

Four areas we would rather quote than pretend about.

These exist in the catalog but not at volume. If your spec lands here, expect three to four weeks to first batch instead of nine days.

  • Construction & trades410 hSite access and PPE compatibility are the bottleneck, not operator supply.
  • Clinical & care180 hCapture only in training environments with simulated patients. No real patient footage, ever.
  • Agriculture120 hSeasonal. Ask about the capture window before you scope against it.
  • Hospitality95 hFront-of-house is bystander-dense, so yield after de-identification is low.

Hours in thin areas are captured to the same spec and sync ceiling as everything else. Thin means fewer cells, not lower quality.

What ships with every clip

Ten streams, one clock, one episode file.

Not an à la carte menu. Every validated hour we deliver carries the whole stack, in the schema below, whether you asked for depth or not.

StreamSpecDetail
Head camera3840 × 2160 · 60 fpsGlobal-shutter, 120° HFOV, rolling-shutter-free, H.265 + lossless keyframes
Wrist cameras × 21920 × 1080 · 60 fpsLeft + right, 100° HFOV, rigid mount, extrinsics re-solved per session
Depth848 × 480 · 30 fpsActive stereo, 0.3–4 m range, metric millimetres, per-frame confidence map
IMU200 Hz · 6-DoFAccel + gyro, bias-calibrated, hardware-timestamped on the same clock domain
Hand pose21 keypoints × 2 hands3D metric, per-joint visibility flag, contact events on grasp and release
Body pose24 joints3D, root-relative and world-frame, torso and forearm chains resolved
SegmentationInstance masksManipulated objects + target surfaces, tracked IDs across the episode
Action segmentsVerb + noun taxonomy97 verbs, 512 nouns, start/end to the frame, human-reviewed
FormatsRLDS · LeRobot · WebDatasetAlso HDF5, zarr, and .rrd for Rerun. Converters shipped as source.
LicenseCommercial · buyer-ownedPerpetual, irrevocable, model-weights-clean. Exclusivity available.

Colour dots map to the modality legend used in every chart on this site. Full field-level schema in the episode schema docs.

Scope a pilot against one named skill.

Fifty to a hundred validated hours, fixed scope, two weeks. Enough to find out whether your architecture responds to the data before anyone signs a coverage program.