Launch your first pilot data program
Before committing to scale, HumanoidLayer runs a bounded, structured pilot batch under your spec. You define the requirements. We deliver QC-reviewed data. You validate the format, quality, and protocol — then decide what comes next.
Why pilot first
What a pilot program gives you
A pilot is not a proof of concept. It is the first production batch of your actual data program — bounded in scope, priced separately, and designed to align on format, quality, and workflow.
Fast time to first data
A pilot batch can be scoped, launched, and delivered faster than building collection ops in-house.
Format and schema validation
You see the exact metadata structure, quality criteria, and delivery format before scaling.
QC calibration
We run the acceptance pipeline against your criteria. You review what passes, what fails, and why.
Two capture modes available
Choose video-based task capture or glove-based hand capture — or request a comparison pilot.
Designed to iterate
After reviewing the pilot batch, we refine the protocol before moving to volume collection.
Buyer-specific from day one
Every pilot starts from your spec — tasks, modalities, volume, and delivery requirements.
Engagement model
How a pilot program unfolds
- 10130–60 min
Intro call
You share the use case, data type, and target deliverable. We confirm whether the program is a good fit and outline a pilot scope.
- 2022–5 days
Spec definition
We jointly define tasks, modalities, QC criteria, volume, timeline, and delivery format. This becomes the production protocol.
- 3031–3 weeks
Pilot program launch
We run the collection program with trained contributors or partner workforces using the agreed setup — video or glove capture.
- 4043–5 days
QC and batch review
We run the acceptance pipeline. You receive a structured batch with QC reports. We review together and align on pass/fail criteria.
- 505Ongoing
Protocol refinement and scale decision
Based on the review, we adjust the protocol. You decide whether to move into recurring collection or an ongoing data program.
What you receive
Pilot batch deliverables
At the end of a pilot, you receive a structured, QC-reviewed data package — not a raw dump.
- Accepted episode batch (as agreed in pilot scope)
- QC report with pass / fail breakdown
- Capture protocol documentation
- Metadata schema and field definitions
- Contributor instruction set (on request)
- Delivery format specification
Setup options
Choose your capture mode
Setup 1 — Video Task Capture
Faster to deploy. Ideal for egocentric task video, workflow capture, and repetitive physical tasks. Lower equipment cost and easier onboarding.
See full setup detailsSetup 2 — Glove Hand Capture
Higher-value signal for hand, finger, and manipulation data. Best for grasp sequences, finger pose, and teleop-compatible datasets.
See full setup detailsFAQ
Pilot program questions
What is the minimum pilot size?
Depends on the task and modality. We typically recommend a pilot of 50–200 episodes to validate format, quality, and workflow before scaling.
How long does a pilot take?
From spec definition to delivered batch, most pilots complete within 2–4 weeks. Complex setups may take longer.
Is a pilot paid?
Yes. Pilot programs are priced separately from volume contracts. Pricing depends on setup, volume, and QC complexity.
Can we run pilots for both capture setups?
Yes. We can run parallel or sequential pilots for video-based and glove-based capture if you want to compare outputs.
What happens after the pilot?
If the batch meets your quality bar, we move into a scaled collection plan. If not, we refine the protocol and run another iteration.
Ready to launch your pilot?
Send us your use case and data requirements. We will propose the right setup, scope, and pilot structure.
hello@humanoidlayer.dev