Stereo Egocentric Data from Construction Crews

It is first-person video from two calibrated cameras, recorded from a crew member’s own viewpoint while they do real jobsite work, with hand pose and sensor streams alongside. Stereo describes the capture and the crew is the subject: the people doing framing, drywall, painting, tiling and the cleanup around them.

This page covers only the 15 rows the site lists under the Active Construction Site environment. It does not claim stereo is already running on any of them. The catalog lists the capture tier for every row as a head-mounted phone or multiview rig, and a stereo pair is a capture choice agreed in the skill spec. Firsthand is not affiliated with Foxglove or the MCAP maintainers.

The Active Construction Site environment lists 15 task rows. Four are marked Ready to deploy (Construction General Skilled Trades, Drywall Hanging, Taping & Mudding, Insulation Installation, and Painting & Priming) and the other eleven are Active collection. By difficulty, six are Hard, seven are Medium and two are Easy. The first table below lists every row with the catalog’s own description.

Several rows are crew-style marking and finishing passes: measure and mark lumber with a tape, tape and mud drywall seams with a taping knife, mask edges and cut in corners with a brush, and grout tile joints. The catalog describes the steps, not a surface-quality or inspection standard. What counts as an acceptable finish is written into the accept criteria of the skill spec.

The hard-environments page states that a stereo pair at human-eye spacing makes depth recoverable from every frame pair, and that stereo keeps working outdoors where active depth sensors struggle, with many programs capturing both and syncing them on one clock. Whether a given jobsite is indoors or outdoors is not something the catalog rows state.

The head camera can lose sight of a grasp when the hand blocks it, which is why wrist views and hand pose travel with the stereo pair. Hand pose is 21 joints per hand with a per-joint visibility flag, so a hidden joint is marked as hidden rather than guessed.

The site describes two stereo setups and does not say which one a construction order uses. The hard-environments page lists a global-shutter 4K/60 head camera with a 120° lens, two 1080p/60 wrist cameras, a depth sensor at 848×480 and 30 fps, and 200 Hz 6-DoF IMUs, with intrinsics and extrinsics re-solved at the start and end of every session. The off-the-shelf stereo head rig, dataset 06, lists dual fisheye cameras at 1080p per eye, a 62 mm baseline, per-unit calibration, a 9-axis IMU and hand-tracking output.

Which rig, resolution and streams a construction order uses is Not stated on the site. It is set in the skill spec and confirmed on a Spec Pilot. The second table below sets what is stated beside what is not.

Capture is planned around the site’s own layout and schedule rather than asking the work to pause. Other trades and crew can end up in frame who are not contributors, so faces and screens are detected and irreversibly blurred at ingest, a reviewer confirms the result, and a clip that cannot be properly de-identified is discarded rather than shipped with a caveat.

The spec names the task once and several contributors are captured performing it, so real variation in technique is represented. Failure and recovery cases, such as a misaligned fastener or a redo of a measuring step, are part of the same labeled structure and can be called out as a requirement for the batch.

Yes. Firsthand offers MCAP delivery, and the MCAP hub page covers it in full, including what the container is according to its own documentation. MCAP is an open container for multiple channels of timestamped, pre-serialized data, so a stereo pair, hand pose and IMU can share one file on one log-time axis.

Topic names, message schemas, encodings and rates are Not stated on the site and are agreed in the skill spec, as is which streams, such as depth, wrist cameras or action segments, go in the file. If your training loop expects RLDS, LeRobot or WebDataset, name that format alongside MCAP.

Off-the-shelf stereo data covers what is already in stock, and the stereo category page lists it. 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.

Construction and skilled trades is listed as a thinner-coverage setting, so a first batch typically takes three to four weeks, not the a median of 48 hours available where coverage is deep. The bottleneck the site states is vetting and scheduling operators with the right site access, not demand for the task.

Name the trade tasks from the rows above, the setting, the streams you want in the file, the delivery format and the accept criteria for what a finished pass looks like. Most orders begin with a Spec Pilot, a small first batch that proves the spec and the accept criteria on real footage before you scale.

Rejected episodes go to a reject log that ships with every batch and are recaptured rather than shipped with a caveat. Pricing is quoted per spec. The fourth table below shows what each spec element decides.