Step 1 of 5
Scope
We go through the tasks, environments, volumes and formats you need, and check them against the work already happening in the network.
You receiveA written scope and a quote.
Real-world training data for humanoid robots.
First-person video of professionals doing real paid work — cleaning, repair, assembly — in customers' homes across Europe, the Gulf, and North America. Not chores staged for a camera.
Whole jobs
Setup, retries, failures, cleanup — the full job, not the clean take.
Diverse and recurring
Thousands of homes across the network, where rare repair and assembly tasks recur.
Consented at source
Consent captured at marketplace, worker, household, and episode level.
From first call to delivery
How a project runs.
Five steps from the first call to data in your training pipeline. A pilot runs through all five at small scale, so what you review is what you scale.
Step 2 of 5
Pilot
A small first batch, collected and packaged exactly the way the full run will be.
You receiveA pilot batch to run through your pipeline.
Step 3 of 5
Review
You check the batch against your own criteria. We adjust capture, labels and packaging until it passes.
You receiveAgreed acceptance criteria for the full run.
Step 4 of 5
Scale
Recurring batches at the agreed volume, from the same network of professionals and homes.
You receiveRegular deliveries, each checked against the pilot criteria.
Step 5 of 5
Deliver
Data lands in the formats you already use — RLDS, WebDataset, HDF5 — with the consent record for every recording.
You receiveA dataset ready for training.
Training a humanoid model? Let's talk data.
Tell us the tasks and formats you need — we'll match them to work already happening.