Egocentric Light Assembly Manufacturing Video Data

Egocentric light assembly data is first-person video of a person assembling, trimming, inspecting or packing light goods, recorded from their own point of view with synchronized sensor streams alongside. The camera sits on the head, close to what a robot’s onboard camera would see, so the frames show what the hands are doing with each part. The same idea sits behind any light assembly robot training data or plastic parts assembly video dataset request, applied here as a custom collection scoped to your tasks.

On this site the data is human capture: a paid, consented contributor wears the rig and performs a task from a written skill spec. It is not robot or teleoperation data. The electronics assembly and screwdriving pages each have their own scope.

The task catalog lists the Plastic & Light-Assembly Manufacturer environment under the Manufacturing & Workshops sector. It has 15 rows, and its example line reads "Consumer goods assembly, packaging operations, labeling, quality inspection on a production line". Every row is Ready to deploy, which the catalog defines as contributors and site access already set up.

The first table below lists the ten assembly, trimming, molding-machine and inspection rows. The second lists the five packing, palletizing and labeling rows, which sit in the same environment. Skill ratings are Easy or Medium; no row is rated Hard.

The site states no machinery models, part or material types, product categories, plant types or sizes, regions, cycle times or line speeds.

Each attempt is delivered as one episode. On this site the standard episode carries action segments, 3D hand pose for both hands with a per-joint visibility flag, contact events that mark when a hand meets an item, an objects list, and per-frame instance masks stored at the depth resolution of 848×480. The capture setup table below lists the stated streams and sync figures.

The catalog lists one capture mode for every light-assembly row: a head-mounted phone with a multiview rig. Force and tactile readings are not among the listed streams.

Action segments divide an episode into start- and end-timestamped spans, each carrying start_ns, end_ns, verb and noun. For line work, a segment might be one part fitted, one flash trimmed or one unit labeled, as written in the skill spec. The site lists no label for the outcome of a step, such as whether a unit passed inspection.

Validation runs against signed accept criteria written into the spec before capture, and a reject log ships with each batch. A human reviewer checks every episode. Only footage that clears the criteria is billed as a validated hour.

Delivery is in the format your stack already reads: RLDS, LeRobot, WebDataset, HDF5, zarr or Rerun .rrd, or raw streams with metadata, with readable converter source alongside. The license is perpetual, irrevocable and worldwide, buyer-owned and non-exclusive by default, with exclusivity available as a separate term. The executed agreement controls.

The catalog describes the work in short, repeated steps: fit the part, fasten it, trim the edge, check the unit, set it in a bin. Fingers cover the small part they are placing, and the part covers the fixture. That is why hand pose carries a per-joint visibility flag, why contact events mark when the hand meets the item, and why masks are recorded per frame. The challenge table below maps each difficulty to the stream or label this site states.

Tolerances, defects and line speed are separate problems. The site lists no tolerance, defect label or line speed for this environment, so the table says so rather than implying a fix.

A robot foundation model paper on arXiv puts human videos at the base of its data pyramid and says they have no explicit action labels, so it learns latent actions. That is robot-model research, not Firsthand data, and Firsthand has no relationship with its authors.

The catalog describes the head-mounted phone as recording the first-person view of real work. The site also states that Firsthand does not capture at a customer’s own facility. Which line or bench a program uses is agreed in the skill spec.

Every contributor signs a written release, and each delivered episode links to it through consent artefact IDs. At ingest, automated detection finds faces, plates and screens and blurs them irreversibly, and speech is scrubbed of names, addresses and card numbers. A human reviewer confirms that no un-consented person is identifiable in any frame, and the anonymization record ships with the data.

Safety certifications, protective equipment, machine guarding and quality-system status are not stated for this environment, and this page makes no such claim. Who may be in frame and what the footage may show are agreed in the spec.

From signed spec to first validated batch, the published timeline is a median of 48 hours where coverage is deep and three to four weeks where it is thin and contributors need to be recruited first. Every light-assembly row is marked Ready to deploy, so contributors and site access are already set up for the catalog tasks. Which range applies is confirmed when the spec is quoted rather than assumed.

Describe the assembly, trimming or inspection steps and their order, the parts, jigs and bins involved, what counts as a finished step, and the failure or variation cases your pipeline is missing. That becomes a written skill spec, quoted for reach, timeline and price before any capture begins. Our skill spec guide walks through how to write one.

Most programs begin with a Spec Pilot, a small, fast first batch that proves the spec and the accept criteria on real footage before volume scales. Collections are billed per validated hour. Counts of sites, episodes and hours are set in the skill spec.