Datalab
Sign inRequest dataEarn
Dataset catalog · illustrative

Data shaped around the model.

Commission multimodal collections with explicit capture specifications, annotation plans, consent coverage, and measurable QA.

Delivery: LeRobot · MCAP · COCO · JSONL
Robotics

Egocentric & RGB-D

Synchronized first-person video, depth, pose, and optional IMU for manipulation and navigation.

Capture
4K RGB · 30–60 fps · depth calibrated
Annotations
Events · objects · actions · quality flags
QA approach
Automated media probes + blinded human sample
Example scale
500+ hours · 12 environments
Explore specification
Vision

Image collections

Consent-backed, diverse image sets with classification, detection, segmentation, and OCR labels.

Capture
RAW/JPEG · EXIF policy · deduped
Annotations
Events · objects · actions · quality flags
QA approach
Automated media probes + blinded human sample
Example scale
8M+ images · 40 languages
Explore specification
Multimodal

Video datasets

Natural workflows, activities, and environments captured against an approved coverage plan.

Capture
1080p–4K · temporal events · audio optional
Annotations
Events · objects · actions · quality flags
QA approach
Automated media probes + blinded human sample
Example scale
22K+ hours · sampled QA
Explore specification
Language

Speech & audio

Prompted, conversational, paired, and environmental audio with isolated tracks and transcripts.

Capture
48 kHz WAV · locale metadata · SNR checks
Annotations
Events · objects · actions · quality flags
QA approach
Automated media probes + blinded human sample
Example scale
70+ locales · human reviewed
Explore specification
Collection capabilities
CaptureManaged network · commercial sites · companion upload
Annotation servicesClassification · detection · segmentation · transcription · temporal events
Quality evidenceCoverage reports · agreement · deduplication · limitations
Delivery formatsLeRobot · MCAP · COCO · JSONL · buyer-defined manifests