Stereo Press-Fit and Snap-Fit Assembly Data

It is real-world, first-person video of a person doing small-part insertion at a precision assembly bench, recorded with two calibrated cameras so that depth can be recovered, together with 3D hand pose and the other streams the site already lists. The person presses a connector onto a PCB, snaps two housing halves together or inserts a peg or dowel into a hole. The buyer sets the task and the accept criteria; Firsthand handles contributors, capture, QA, consent and delivery.

The data is human capture, not robot or teleoperation data, and it is not a fixed corpus you download. The same request is also phrased as press-fit assembly robot training data, stereo snap-fit housing video, an egocentric peg-in-hole insertion dataset or custom precision assembly data collection at scale: write the task down, collect it on demand and receive validated episodes.

Three live rows in the Precision Assembly Bench (electronics) environment name this work. "Press-fit connectors" reads "Place and press connectors onto a PCB." "Snap-fit housing" reads "Snap two housing halves together." "Peg-in-hole insert" reads "Insert pegs/dowels into holes." All three are rated Medium, list a capture setup of Multiview rig + handheld grippers, and are Available on request. The first table below reproduces them exactly.

The environment sits in the Manufacturing & Workshops sector and holds fifteen rows, all Available on request. The catalog describes it as "Assembly bench: PCB/electronics, fasteners, fixtures, solder/crimp tools, small parts". A fourth insertion row, "Place into fixture", is covered on the egocentric peg-in-hole page and is not claimed here. This page claims no delivered hours or delivered stereo footage for any row. Part types, forces, tolerances, fit classes, cycle times and error rates are Not stated, so they are set in the written spec.

Not stated. The catalog lists a multiview rig with handheld grippers for all three rows and no stereo pair. Stereo applies here only through the site general stereo offer: the site describes a stereo head rig and a hard-environments stack, and a stereo pair for a press-fit or snap-fit order is a choice made in the skill spec and confirmed on a Spec Pilot.

That is why the page is written as a stereo and container page for the three rows rather than as a description of existing stereo insertion footage. If you need stereo, say so in the brief. If the catalog setup is enough, the egocentric peg-in-hole and precision bench pages describe that non-stereo route.

This page is stereo capture of press-fit, snap-fit and peg-in-hole assembly on a precision assembly bench. The egocentric peg-in-hole page covers four insertion rows without a stereo pair, and the egocentric precision bench page covers all fifteen bench rows, also without stereo. Both are Available on request.

Three other stereo pages sit nearby: consumer goods assembly (three Plastic & Light-Assembly rows, listed as Ready to deploy), torque and bolt fastening (two Medium rows) and wire crimping and harness assembly (one Hard and one Medium row). The second table below sets them side by side with their status as the site lists it.

The site describes two stereo configurations. The stereo head rig, dataset 06, is a dual fisheye pair at 1080p per eye on a 62 mm baseline, with per-unit calibration, a 9-axis IMU and hand-tracking output, shipped as MCAP. The hard-environments stack is a 4K/60 head camera with a 120° lens, two 1080p/60 wrist cameras, depth at 848×480 and 30 fps and a 200 Hz 6-DoF IMU. Which of these a press-fit or snap-fit order uses is Not stated and is agreed in the skill spec.

Whichever rig is chosen, calibration is measured every session and shipped with the episode, with a 0.28 px RMS reprojection gate. The catalog setup for these rows pairs a multiview rig with handheld grippers, which the site describes as letting demonstrations map onto a robot end effector; no accuracy figure for that mapping is stated. The language data is one task instruction per episode.

Small parts are the first case. The site states that small fasteners can fall below a depth sensor floor, that hand pose stays valid and that sub-centimetre metric depth is flagged rather than interpolated. It states no depth floor for pegs, pins or housing features, so very small parts are confirmed in the spec.

Occlusion at the moment of insertion is the second. The hand covers the part and the part covers the seat, so the per-joint visibility flag marks joints that are not visible instead of guessing them, contact events mark when a hand meets an item and instance masks are recorded per frame. Glare is the third. The site states that depth degrades on wet and reflective surfaces and that per-frame confidence maps mask low-confidence regions. Lighting values are Not stated and are fixed in the spec.

Each attempt is delivered as one episode. The site episode schema lists action segments, contact events, 3D hand pose for both hands with a per-joint visibility flag and instance masks, alongside int64 nanosecond timestamps per stream and camera intrinsics and extrinsics. Action segments carry start, end, verb and noun, so one press, snap or insertion can be its own segment as the spec defines it.

Validation runs against accept criteria written into the spec before capture, and a reject log ships with each batch. The site states these reject thresholds: a cross-stream sync error above a 2.0 ms ceiling, the manipulated object out of frame for more than 15 percent of the episode, and more than 35 percent invalid depth pixels in the manipulation region. They are thresholds, not a pass rate. Rejects are recaptured at the provider cost. Contact events are not force or pressure.

Firsthand offers MCAP delivery. MCAP is an open-source container format with its own official site at mcap.dev, a specification at mcap.dev/spec and a repository at github.com/foxglove/mcap; read those for what the container is and how it stores records, timestamps and attachments. This page describes the container only from those sources and does not state anything more about it.

The streams the site already states for a stereo capture are the stereo pair, depth with confidence maps, 21-joint hand pose, IMU, contact events, action segments and calibration. Topic names, message schemas, encoding and rates are Not stated on the site and are agreed in the skill spec. Only dataset 06 is stated as shipped as MCAP; for a custom order, MCAP can be the only delivery or sit beside RLDS, LeRobot, WebDataset, HDF5 or zarr.

Peg and hole dimensions, fit classes, clearances, insertion or press forces, connector, housing and fixture types, counts, cycle times and task times are not stated for these rows. Force, tactile and torque readings are not among the standard streams; standard capture is cameras, depth and IMU.

The site lists no seated, snapped, jammed or defect label and no pass or fail label. This page gives no accuracy, success rate, benchmark or robot result, and names no client, plant, brand or product type. Anything you need beyond the stated streams is written into the skill spec and covered in the quote.

No off-the-shelf dataset for press-fit, snap-fit or peg-in-hole tasks exists on the site, so this is custom collection. The off-the-shelf stereo category lists stereo datasets and hand-tracking video, and only dataset 06 is stated as MCAP. It is a way to look at the format and rig, not coverage of these three rows. Custom collection covers the rest: custom collection at scale through a network of thousands of capture providers, for a consistent, recurring data flow to labs.

Where coverage is deep, the median time from signed spec to first validated batch is a median of 48 hours; where it is thin, turnaround runs three to four weeks. Which range applies to this bench is confirmed in the quote. Collections are billed per validated hour.

The site states that faces and screens are blurred at ingest, and describes consent and anonymization on its anonymization page. Anything beyond that, including how a venue handles other people or marked parts in frame, is Not stated and is set in the spec. This page gives no equipment-safety guidance.

The buyer owns the dataset under a perpetual, irrevocable, non-exclusive-by-default licence, with exclusivity available. Consent is recorded per session.

Name the row you are following, what is pressed, snapped or inserted, what counts as one finished step and what a reject looks like. Then name the streams, the topic naming and message encoding your reader expects, and whether MCAP is the only delivery.

Most programs begin with a Spec Pilot, a small first batch that proves the spec on real footage before you scale. The last table below shows what each spec element decides.