Stereo Concrete Finishing Data

It is first-person video of a person doing concrete work, recorded from their own point of view with two calibrated cameras and with hand pose and sensor streams alongside. The task is the one the site’s catalog lists: mix, pour, screed and finish concrete into forms using shovels, floats and trowels. Stereo describes the capture, and MCAP describes the container. They are separate choices.

The site does not state that stereo is already running on the concrete row. The catalog lists a head-mounted phone or a multiview rig as the capture setup, and a stereo pair is a capture choice agreed in the skill spec and confirmed on a Spec Pilot. The data is human capture by paid, consented contributors, not robot or teleoperation data. Firsthand is not affiliated with Foxglove or the MCAP maintainers.

The surface finishing page covers nine rows across sanding, drywall and painting. The crews page covers the 15 rows on the Active Construction Site environment. The drywall and wall finishing page covers taping, mudding and wall finishing as a task. None of them covers the concrete row, which lives on a different environment, Heavy Industry & Fabrication, and which is rated Hard and Available on request rather than Ready to deploy.

This page covers only that row and the one general trades row that names concrete. It does not repeat the crews or surface finishing material. It covers the two choices layered on top of the concrete rows: a calibrated stereo pair for depth, and MCAP as the delivery container. The masonry and plastering page mentions the same row as context and makes no structural claim, and this page does the same.

Two rows name concrete, reproduced exactly in the first table below. "Concrete work" sits in the Heavy Industry & Fabrication environment, in the Manufacturing & Workshops sector: "Mix, pour, screed and finish concrete into forms using shovels, floats and trowels." It is rated Hard, lists "Phone (head-mounted) / multiview rig" as its capture setup, and is Available on request.

"Construction General Skilled Trades" sits in Active Construction Site, in the Construction & Trades sector: "Skilled trade work on active sites including framing, concrete and structural tasks." It is rated Hard and Ready to deploy. It is a general trades row that names concrete among other work, not a concrete finishing row, so this page uses it as context for where concrete work can sit on an active site and makes no finishing claim from it.

Heavy Industry & Fabrication holds five rows, all Hard and all Available on request: concrete work, glass manufacturing, metal fabrication, operating heavy machinery controls and welding. The heavy machinery row is covered on its own page, and it is the operator-station counterpart to this hand-tool row. The catalog does not name edging, jointing, brooming or curing, and it does not name a power trowel or a bull float. Anything beyond the row’s own description is a skill-spec question for a Spec Pilot.

Available on request means the work is scoped and recruited to your spec when you ask, rather than already queued. For the concrete row that means contributors with the right access are found after the spec is written. Ready to deploy means contributors and site access are set up and capture starts once a spec is signed, which is the status of the general trades row.

The site does not say whether coverage for the concrete row is deep or thin, so ask when you request the quote. 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. The crews page states that construction and skilled trades is a thinner-coverage setting, so for the general trades row on an active site the three-to-four-week figure is the one that applies.

A lab building a finishing policy needs the pass structure, the tool in the hand and the state of the surface. The pass structure is the ordered run the row describes: mix, pour, screed, finish. The site’s delivered fields include action segments, so each sub-step can be a labeled start and end, and the skill spec decides what one pass is and where an episode starts and stops.

The tool in the hand is hand pose with a per-joint visibility flag, 21 joints per hand, alongside contact events that mark when a hand meets the tool or the item. The state of the surface is what stereo adds: depth recovered from the calibrated pair, with a confidence map that masks regions the pair cannot be trusted on. The site states no surface-quality, flatness or finish-grade label for these rows, so any such label is a buyer-side or spec-side decision and is Not stated here.

A lab also needs to know what the data is not. It is human capture of hand-tool work, so it records how a person moves a float or trowel, not robot joint commands or forces. The site does not describe a force stream for the concrete row.

Finishing is about the state of a surface: whether it is level, whether it has been worked, where the edge of the pass lies. A stereo pair recovers depth from the calibrated frame pair rather than from a standalone depth sensor, so the form of the slab and the forms around it sit in the same frame as the hands. The site gives no depth accuracy figure for this row and no flatness label, so what a buyer does with that depth is their pipeline’s question.

The hard-environments page states that stereo keeps working outdoors where active depth sensors struggle, and that many programs capture both and sync them on one clock. Whether a concrete setting is indoors, outdoors, a plant or a site is not something the catalog row states, because the row’s examples list a concrete plant among other heavy industry settings and the general trades row sits on an active site. That is agreed in the spec.

The site describes two stereo setups and does not say which one a concrete order uses. The hard-environments stack is 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. The off-the-shelf stereo head rig, dataset 06, is 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 concrete order uses is Not stated on the site, and the second table below sets what is stated beside what is not. Intrinsics and extrinsics are re-solved at the start and end of every session, and streams are hardware-triggered onto one clock with a 2.0 ms cross-stream ceiling. Wrist views can be added in the spec, which matters for a grip the head camera loses behind a float or a trowel.

Wet surfaces are the first risk. The site says depth degrades on wet and reflective surfaces, and a freshly worked surface is wet. Confidence maps mask the low-confidence regions, and an episode is rejected when more than 35 percent of depth pixels in the manipulation region are invalid. The site does not say how depth behaves on fresh concrete in particular, so this page promises no figure and no usable-depth percentage.

Hands are the second. A float or trowel sits flat on the surface and the fingers wrap the handle, so the contact point is often hidden, and a shovel grip can be hidden by the load. The visibility flag on each joint is the answer the site gives, and contact events mark when a hand meets the tool. Wrist views can be requested so a grip the head camera loses is still recorded.

Calibration is the third, because stereo depth is only as good as the rectification behind it: calibration is measured at the start and end of every session under a 0.28 px RMS gate. An episode is also rejected when the manipulated object is out of frame for more than 15 percent of it or an undocumented bystander appears, and a reject log ships with every batch. Other people on a shared site can end up in frame, 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.

MCAP is an open source container file format for multimodal log data, supporting multiple channels of timestamped pre-serialized data. It is serialization-agnostic, so messages can be Protobuf, DDS (CDR), ROS, JSON and more. Its design is row-oriented and append-only, it stores schemas alongside the data, and it offers an optional index plus optional LZ4 or Zstandard chunk compression.

Reader and writer libraries exist for C++, Go, Python, Rust, Swift and TypeScript, and the mcap CLI can inspect, merge and split files. The foxglove/mcap repository is MIT-licensed; that is the MCAP project’s licence, not the licence of a Firsthand delivery. The MCAP hub page and the glossary entry cover the format in full.

Firsthand offers MCAP delivery, and dataset 06 is stated as shipping as MCAP. For finishing work the site states hand pose, a stereo-derived depth stream, IMU, contact events, action segments and per-frame instance masks elsewhere. What it does not state is which of those are channels in an MCAP file, how they are named, what encoding they use or at what rate.

No topic name, message schema or rate is stated on this page, and none should be assumed. Name the streams that matter in the skill spec and confirm them on a Spec Pilot batch. If your training loop expects RLDS, LeRobot or WebDataset, name that format beside MCAP. The third table below sets each stream beside what the site states about it.

Off-the-shelf covers what is already in stock: the stereo category lists the stereo datasets and hand-tracking video, and only dataset 06 is stated as MCAP. This page does not claim off-the-shelf coverage of concrete finishing. 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.

A lab can use the stock stereo and hand-tracking video to start building a pipeline, then order concrete-specific batches against one written spec so each recurring batch is captured and checked the same way. Collections are billed per validated hour, and capture at your own facility is listed as not offered. The site states no batch size, cadence, provider count or hours for concrete, so those are Not stated.

Describe the steps and their order, what counts as a completed pass, the surface and the setting, and the variation your pipeline is missing. Then name the streams you want in the file, the topic naming and message encoding your reader expects, whether calibration ships inside the file or beside it, and whether MCAP is the only delivery or sits beside RLDS, LeRobot, WebDataset, HDF5 or zarr.

Most programs begin with a Spec Pilot, a small first batch that proves the spec and accept criteria on real footage before you scale. Rejected episodes go to a reject log and are recaptured rather than shipped with a caveat. The fourth table below shows what each spec element decides.