Manufacturing & workshop

Custom Assembly Task Data for Robot Learning

Fastening, measuring, press-fit and wire work captured bench-side, at hand-pose precision, against a spec you write rather than a public benchmark category.
Streams
Ten, synced under 2 ms
First batch
Median 48 h
Billing
Per validated hour
License
Buyer-owned
A garage workshop bench with a vise, hand tools on a pegboard and real clutter
Bench assembly is the cleanest environment for bimanual precision because the operator controls the whole scene.

Requirements

What does assembly task data need?

01

Hand-level precision, not just video

Driver and bit selection, clamp-and-align, press-fit, and wire strip-crimp-terminate all live in finger-level detail that RGB video alone cannot resolve. Each needs 3D hand pose tracked through reach, grasp, and place.

02

A named assembly skill list

Fastening, measuring and marking to tolerance, press-fit and interference assembly, and tool hand-off between hands each need their own coverage, not one generic "assembly" label.

03

Depth with known limits on small parts

Small fasteners can fall below a depth sensor’s reliable floor. Hand pose stays valid at that range, but metric depth on sub-centimetre parts is flagged rather than interpolated, so you know exactly where the depth stream stops being trustworthy.

04

Rights you can ship on

A perpetual, buyer-owned commercial license backed by a documented consent chain, so a policy trained on the data can go into a product.

Why it matters

Why is assembly data hard to capture well?

Public benchmark datasets for bimanual precision, such as EgoDex, have genuinely strong hand pose, but their licenses forbid commercial training use and forbid derivatives — so they cannot legally reach production weights.

Bench assembly is also the cleanest environment to run this kind of capture in, because the operator controls the whole scene: lighting, clutter, and tool layout can be varied deliberately rather than relying on incidental footage that was never captured against a named skill list.

Capture setup

How is assembly task data captured egocentrically?

A head-mounted camera keeps the fastening, measuring, and fitting motion in frame from the viewpoint a manufacturing policy will actually see from, across home workshop, maker bench, and small production cell environments.

Depth, 3D hand pose, contact events, and action segments are aligned to video on one hardware clock. Small fasteners below the depth sensor floor are flagged rather than interpolated, while hand pose remains valid at that range.

The spec

How does a custom assembly spec work?

You describe the assembly tasks your policy needs. We turn it into a written spec covering tasks, environments, capture tier, annotation depth, and accept criteria, then quote reach, timeline, and price against that document before any capture begins.

The collection runs through the same seven-stage end-to-end custom collection pipeline as every Firsthand program: contributor recruiting, capture, per-batch QA, anonymization, and licensed delivery. Rejected items are recaptured at our cost.

Fast · on demand

How fast does a first batch ship?

Where coverage is already deep, the first validated batch lands in a median of 48 hours from signed spec. Thin environments typically take three to four weeks, because vetting operators there is the bottleneck, and we quote that timeline up front.

On-demand top-ups run against the same spec and accept criteria after each training run, so new assembly data stays consistent with what you already hold.

Rights

What do licensing and consent look like?

Every custom collection ships under a perpetual, buyer-owned commercial license, and exclusivity is available so the data is never resold. Each episode links to a signed contributor release through a documented consent chain.

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.

Assembly task vs delivery

What does an assembly policy get, task by task?

Fastening, measuring, and fitting tasks and what a Firsthand custom collection delivers for each.
Assembly needWhat Firsthand captures
Bench work in frameHead-mounted camera keeping fastening, measuring, and fitting motion visible through the task cycle
Named skill coverageDriver and bit selection, clamp-and-align, measure and mark to tolerance, press-fit, wire strip-crimp-terminate, tool hand-off
Hand-level detail3D hand pose tracked through reach, grasp, and place
Contact detailContact and grasp events at the point of fastening or fitting
Depth on small partsSynchronized depth stream, with metric depth on sub-centimetre parts flagged rather than interpolated
Action structureInstance segmentation, action segments, and frame-accurate task boundaries
QualityMeasured against signed accept criteria; billed per validated hour; reject log with every batch
FormatsRLDS, LeRobot, WebDataset, plus HDF5, zarr, and Rerun .rrd

Every collection runs through the same seven-stage end-to-end custom collection pipeline.

FAQ

Assembly task data, answered.

What is assembly task data?
It is real-world, first-person demonstration data of manufacturing and workshop tasks — fastening, measuring, press-fit, and wire termination — captured against a written spec: egocentric video with depth, 3D hand pose, contact events, and action segments. Firsthand handles contributors, capture, QA, consent, and delivery.
How is this different from a public dataset like EgoDex?
EgoDex has strong hand pose but its CC BY-NC-ND license forbids commercial use and forbids derivatives, so it cannot legally reach production weights. Custom assembly task data is buyer-owned and perpetual, captured against your own named skill list.
Can depth be trusted on small fasteners?
Hand pose remains valid at close range, but metric depth on sub-centimetre parts falls below the depth sensor’s reliable floor and is flagged rather than interpolated, so you know exactly where to stop trusting that stream.
How is custom assembly data billed?
Per validated hour, quoted against the written spec before any capture. You pay only for data that clears the accept criteria, and rejected items are recaptured at our cost.
Which delivery formats are supported?
RLDS, LeRobot, and WebDataset, plus HDF5, zarr, and Rerun .rrd. Converter source is readable, so you can retarget the schema to your own loader.
Can we start with a small pilot?
Yes. A Spec Pilot is a small, fast first batch that proves the spec and accept criteria before you scale volume on demand.

Sources

Where are the delivery formats documented?

Request a custom dataset.

Tell us the assembly tasks your policy needs to perform. We will quote reach, timeline, and price against a written spec, with a first validated batch in a median of 48 hours where coverage is deep.