Egocentric Thermal Paste and Heatsink Application Data

It is first-person video of one person applying thermal paste and seating a heatsink, recorded from that person’s own point of view. A paid, consented contributor performs the task written in a skill spec, and each attempt is delivered as an episode with synchronized streams, hand pose and labeled action segments. It is human capture, not robot or teleoperation data, and not a fixed corpus you download.

The catalog row is written in two parts: "Apply thermal paste and seat a heatsink." This page keeps to the row as written. It names no paste, heatsink, board or tool, because the site names none, and treats anything more specific as a spec question settled in a Spec Pilot. It makes no claim about a trained model, benchmark or success rate.

One row counts. "Apply paste + heatsink" sits in the Precision Assembly Bench (electronics) environment, in the Manufacturing & Workshops sector, whose example line reads "Assembly bench: PCB/electronics, fasteners, fixtures, solder/crimp tools, small parts". The row is described as "Apply thermal paste and seat a heatsink.", rated Medium, listed with "Multiview rig + handheld grippers" and Available on request.

Available on request is the site’s wording for a task scoped and recruited to spec when a buyer asks, as opposed to Active collection or Ready to deploy. So this page claims no delivered hours, stock of footage or published dataset for this task. The first table below reproduces the row, with a neighbouring bench row as context only.

The row says what is done, not what it is done with. It names no paste, heatsink, board, tool or part count, and gives no amount, spread pattern, seating order or tolerance. Whether fastening is part of an attempt is not stated. A buyer who needs a pattern, a fastening step, paste only or seating only writes it into the skill spec as an accept criterion, and a Spec Pilot proves it before you scale.

The Precision Assembly Bench lists 15 rows, all Available on request, each with a multiview rig and handheld grippers. The ratings are two Easy, ten Medium and three Hard; the paste row is one of the ten Medium rows.

"Apply adhesive bead" is mentioned as context only. It is described as "Dispense an adhesive bead on a path.", also rated Medium and Available on request. The rest of the bench covers insertion, fastening, fixture loading, connector seating, harness routing, kitting and labeling, and none of those rows names paste or a heatsink. If your project needs several bench tasks, the precision assembly bench page treats all 15 rows; this page takes the paste row on its own.

Fingers cover the small item they are placing, and the item covers the surface it is going onto. Three consequences follow, each already addressed on the site.

Occlusion: fingers disappear behind objects, which is why 3D hand pose carries a per-joint visibility flag. A hidden joint is marked as hidden rather than presented as a guess. Depth: the site states that small fasteners can fall below a depth sensor’s reliable floor, that hand pose stays valid at that range, and that metric depth on sub-centimetre parts is flagged rather than interpolated. No depth floor is stated for this row, so ask for it to be confirmed in the spec.

What video cannot show: the site lists no label for the outcome, and force, tactile and temperature readings are not among the standard streams.

The GR00T N1 paper says its human video datasets do not include explicit action labels, and that hand-held robot grippers speed up collection but leave a mismatch with the downstream robot. Firsthand captures people, not robots, so using the episodes for a specific robot means your own retargeting step.

For "Apply paste + heatsink" the catalog lists "Multiview rig + handheld grippers": a staged working set recorded by a multiview rig while the operator performs the task with handheld grippers. The catalog lists no stereo pair for this row, so this page does not describe the capture as stereo.

The wider stack the site documents is ten streams, hardware-triggered onto one PTP clock domain and aligned under 2 ms, with the rig calibrated at the start and end of every session. Which streams apply to a given paste and heatsink collection, how many cameras are used and what the grippers are is set in the skill spec; the site states no camera count, resolution or frame rate specific to this row.

Each attempt is delivered as one episode, and this row needs a decision up front about what one attempt is: paste only, seating only, or both in sequence. The site does not set that length, so the spec does.

Action segments divide the recording into spans carrying start_ns, end_ns, verb and noun, from a taxonomy of 97 verbs and 512 nouns, annotated to the frame and human-reviewed. A spec that says only "apply paste and seat the heatsink" gives one long segment; one that names the steps gives a sequence you can slice. Other fields are hand_joints and hand_visible (21 joints per hand), contact_events, objects, and instance_masks at 848×480, plus int64 nanosecond timestamps, intrinsics and extrinsics per stream.

The site lists no label for the result: no paste-coverage, seated-state, defect, amount, pattern or completion label. If your pipeline needs one, it is scoped in the spec rather than assumed.

Several pages touch this row, but none takes it as its main topic, which is why this page exists. The comparison table below states each page as the site states it. For one dispensing row against another, start with the adhesive and sealant page; for the whole bench, start with the precision assembly bench page.

Only what the site already states applies. Every participant signs a written release before capture, and consent artefact IDs link each episode to it. Faces, plates and screens are detected and irreversibly blurred at ingest, speech is scrubbed of names, addresses and card numbers, and a human reviewer confirms no un-consented person is identifiable in any frame. Footage that cannot be de-identified is discarded, and the anonymization record is delivered with the data.

Printed text or markings on a board or part are not a listed blur category, so this page does not claim they are blurred. Which bench, venue or region a program uses, and any safety practice, is not stated and is agreed in the spec. This is not legal advice, and the executed agreement controls.

The site states one published timeline from a signed spec to a first validated batch: a median of 48 hours where coverage is deep, and typically three to four weeks where it is thin. It does not say which applies to the Precision Assembly Bench, so ask when you request the quote. A tight spec is the quickest route to a fast, on-demand dataset.

The license is perpetual, irrevocable and worldwide, buyer-owned and non-exclusive by default, with exclusivity available as a separate term. Collections are billed per validated hour, rejected episodes and recaptures are never billed, and the site states no price list for this task.