Buy Robot Pretraining Data

Buying robot pretraining data means commissioning a custom collection of first-person human behavior — egocentric video, depth, hand pose, and action labels — captured to a written spec and delivered under a commercial license, instead of downloading one of the public egocentric corpora referenced throughout this site.

That distinction matters because pretraining data sits at the base of a robot foundation model’s training, per NVIDIA’s GR00T N1 data pyramid: web data and human video at the base because they are most abundant, synthetic data in the middle, and real robot trajectories at the top because they are scarcest and most expensive. A purchase at this layer is judged on breadth across tasks, objects, and environments, and on whether the license actually clears a commercial training run.

Every custom collection ships the same capture stack: ten synchronized streams per episode — 4K head camera, two wrist cameras, depth, IMU, 3D hand and body pose, instance segmentation, and action labels — hardware-aligned under 2 ms. Because hand pose, depth, and action labels are captured live on the rig rather than estimated from raw video afterward, every episode carries a manipulation-grade signal, not RGB alone.

Coverage is set by the written spec you provide — tasks, environments, and regions — rather than a fixed catalog of available footage. That is the core difference from downloading a public corpus: a public dataset was captured for someone else’s task distribution, so it is always missing exactly what a given model needs next, while a custom purchase is scoped to fill that specific gap.

You describe the skill and environment diversity your pretraining corpus is missing. That becomes a written spec covering tasks, environments, regions, capture tier, annotation depth, and accept criteria, quoted for reach, timeline, and price before any capture begins.

Many programs start with a Spec Pilot: a small, fast first batch that proves the spec and accept criteria before you commit to scale. It lets you check stream quality, hand-pose and action-label fidelity, and format compatibility with your training pipeline before volume ramps and before a larger purchase commitment is made.

From there, the purchase scales through the same seven-stage end-to-end pipeline used for every Firsthand program: contributor recruiting, capture, per-batch QA, anonymization, and licensed delivery, with rejected items recaptured at Firsthand’s cost. A median 48-hour turnaround on a first batch means you can validate the spec quickly, then scale the same purchase to the volume your model needs, with on-demand top-ups available afterward as new gaps show up during training.

Public egocentric corpora differ mainly in how they were captured and what they license, not in whether scale is achievable. EgoScale’s action labels, for instance, come from off-the-shelf SLAM and hand-pose estimation applied after footage is captured, and most large egocentric corpora carry non-commercial or research-only terms rather than a right to train a commercial product on them.

A custom purchase closes three gaps a public download cannot: capture method (hardware-synchronized streams and manipulation-grade pose measured live, not estimated after the fact), licensing (a commercial-training license written from the outset, not a research grant), and coverage (a corpus scoped to the exact tasks, objects, and environments your model is missing, rather than whatever someone else happened to record).

None of this makes public pretraining corpora unusable. EgoScale’s own results show a research-grade corpus, action-labeled by off-the-shelf SLAM and pose estimation, is enough to demonstrate a real scaling law and a measurable success-rate gain. The choice is about what a given program needs next: broad, general priors are exactly what large public corpora are built for, while filling a specific, known gap in coverage or clearing commercial licensing terms is what a custom purchase is built for.

Every custom collection ships under a perpetual, buyer-owned commercial license, with exclusivity available so the data is never resold, and each episode links to a signed contributor release through a documented consent chain — the piece many public pretraining corpora do not clear for commercial use.

Faces, plates, and screens are detected and irreversibly blurred at ingest, and a reviewer confirms the result before anything ships. The anonymization record is delivered with the data, so a purchase has a clear paper trail from contributor consent through anonymization to the license terms governing commercial training on the finished corpus.

RLDS, LeRobot, and WebDataset, plus HDF5, zarr, and Rerun .rrd, so a purchased corpus loads directly into the same pipelines used for mid-training and fine-tuning data. Converter source is readable, so you can retarget the schema to your own loader.

Because pretraining, mid-training, and fine-tuning data all ship through the same delivery formats, a purchase does not need a separate ingestion pipeline for each training stage, whether the corpus is ten thousand episodes deep or a small aligned mid-training batch.

There is no fixed price list, because the cost is set by what your spec asks for, not a catalog item. The main levers are how many synchronized streams each episode needs, how many validated (post-QA) hours you need, how thin or deep coverage already is for the tasks and environments you want, and how fast you need the first batch.

Billing is per validated hour: footage that has cleared the accept criteria written into your spec. A take that fails QA is not billed — it is rejected and recaptured, at Firsthand’s cost, until it clears the bar or the spec is revisited. See the full cost-driver breakdown for how each lever moves the quote.

Start with the gap in your current pretraining corpus rather than the total volume you think you need: which tasks, objects, or environments is the model still weak on. That gap becomes the written spec — tasks, environments, regions, capture tier, annotation depth, and accept criteria — quoted for reach, timeline, and price up front.

A Spec Pilot proves that spec on a small, fast first batch before any commitment to scale, so you can check stream quality, hand-pose and action-label fidelity, and format compatibility with your training pipeline before volume ramps.