Measured, certified capture in the in-house POD
Where the marketplace gives scale, the POD gives precision — calibrated ego + exo capture producing measured channels and signed provenance, not estimated labels.
Some training needs channels you cannot recover after the fact — metric-scale pose, calibrated depth, consistent multi-view geometry. Phone footage estimates these; certain buyers need them measured, with a rights trail they can defend under audit.
In our own facility, an operator wears an ego device while synchronized exo cameras surround the workspace, all calibrated into one metric frame. Each session ships measured channels — RGB, 6DoF pose, depth, 3D body and 3D hands — and a C2PA-style signed provenance manifest recording device, pipeline version and consent terms.
- 01
Calibrate
ChArUco calibration solves the exo cameras into the ego frame for one metric world.
- 02
Script tasks
Define the manipulation and activity tasks, scoped to the buyer’s spec.
- 03
Capture measured
Sessions produce RGB, 6DoF pose, metric depth, and 3D body/hands channels.
- 04
Process & certify
The pipeline outputs measured channels and a signed provenance manifest per session.
- 05
Deliver to spec
Datasets are delivered in the buyer’s format with consent attestations attached.
Buyers train against measured pose and depth directly, rather than estimating and correcting downstream. The POD is the half of the supply chain a pure marketplace cannot offer — calibrated, certified, and repeatable — and it shares one consent and provenance standard with the marketplace.
Specs are illustrative of a representative engagement, not a fixed product sheet — every campaign and POD capture is scoped to the buyer.
Related case studies
First-person household tasks for home robotics
Egocentric video of real home and kitchen tasks — the manipulation data home-robotics and assistive-agent teams cite as their biggest bottleneck.
Read case studyRare-environment first-person video for long-tail robustness
Egocentric footage from unusual places and edge-case activities — the long-tail coverage that makes physical-AI models robust where common data can’t.
Read case studyAnimal-enclosure video for livestock and behavior AI
Footage from real pens, barns and habitats — training data for livestock monitoring, agriculture automation, and animal-behavior recognition.
Read case studyQuestions, answered plainly
Have a dataset to build, or footage to earn from?
Buyers: tell us what you're training and we'll scope the supply — marketplace, in-house POD, or both. Creators: browse funded campaigns and get paid for your content and your time.