Egocentric training data for Physical AI
Captured, annotated, and delivered to your pipeline — manipulation, locomotion, and human activity from a first-person view.
We’re onboarding our first cohort of robotics teams. Tell us what you’re training and we’ll build the dataset with you.
Environments
6+
Kitchen · Warehouse · Manufacturing · Outdoor · Lab · Custom
Annotation layers
5+
Depth · Pose · Segmentation · Contact · Metadata
Output formats
Native
LeRobot · RLDS · Open X-Embodiment · Custom
Compliance
Global
Rights-cleared · PII-redacted · GDPR-aligned · Multi-jurisdiction
Data diversity
Every environment your model needs to learn from
Egocentric captures across the full task space — scoped to your use case.
Manipulation
Available on requestObject grasping, tabletop tasks, pick-and-place, tool use.
Request this datasetAnnotation
More than video
Every clip ships with RGB base capture plus five structured annotation layers — ready for your training pipeline, no preprocessing required.
Formats
Delivered to your pipeline
Every dataset ships in your format — no conversion step, no preprocessing overhead.
Hugging Face ecosystem · Direct dataset loading
TensorFlow Datasets · Reverb-compatible
Cross-embodiment training standard
Parquet · HDF5 · zarr · raw NPY
How it works
From brief to delivered dataset
Brief
Tell us what you're training — task type, environment, modality, volume. We confirm scope and timeline in one conversation.
Capture & Annotate
Our collector network captures real-world demonstrations. Every clip is annotated across all layers and QA-verified before leaving the pipeline.
Delivery
Training-ready datasets in your format — LeRobot, RLDS, or custom. Provenance report and QA summary included.
Licensed, real-world. Not synthetic. Not scraped.
Rights-cleared globally
Every clip captured under explicit consent. Multi-jurisdiction rights clearance — not region-locked. GDPR-aligned from day one.
Per-dataset QA
Every dataset includes a QA report before delivery. Annotation accuracy reviewed by human reviewers. Reject threshold enforced at the clip level.
Full audit trail
Complete provenance per dataset — who, where, when. Inspection-ready documentation. Chain-of-custody for every captured sequence.
PII-redacted by pipeline
Automated PII detection runs on every clip — faces, license plates, incidental bystanders — with redaction applied before annotation. Privacy-safe data by process, not by promise.
For data collectors
Your data trains AI. Get paid when it does.
We’re building a network of data collectors across environments. Join the waitlist — we’ll reach out when collection in your area begins.
Built for frontier robotics teams.
Accepting first cohort applications
Tell us what you’re training.
We’ll scope the dataset with you — no long procurement cycle.