Stereo Tile and Flooring Installation Data

It is first-person video of a person laying tile, 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 for tiling: spread adhesive, set tiles with spacers, cut edge tiles and grout the joints. 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 tiling 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 egocentric tile and flooring installation page covers the task: the catalog row, the labels, the privacy steps and what that page does not state. The stereo construction crew page covers all 15 Active Construction Site rows with the crew as the subject, and tiling is one line in its table. The stereo surface finishing page covers rows across sanding, drywall and painting and does not name the tiling row, and the stereo concrete finishing page covers concrete work. This page does not repeat any of them.

It narrows to the one tiling row and asks what a construction-finishing or flooring-robotics lab needs from a stereo capture of it: the geometry of the surface and its edges, the hand on the tile and the tool, the failure modes of a wet bed and a wet joint, and the container the data arrives in. The row is the same as on the sibling pages and the claims are as limited. This page states no tile size, adhesive or grout product, cure time or flatness tolerance, because the site states none.

One row counts. In the Active Construction Site environment the catalog lists "Tiling", described as "Spread adhesive, set tiles with spacers, cut edge tiles and grout the joints." It is rated Hard, its status is Active collection, and its capture setup is a head-mounted phone or a multiview rig. The environment holds 15 rows: six Hard, seven Medium and two Easy, with eleven Active collection and four Ready to deploy. The first table below reproduces the row as the catalog states it.

Two other rows word tiling and are not counted. "Mosaic tiling" sits in the Glass & Mosaic Studio as studio craft, rated Medium and Available on request, and has its own page. "Hotel Bathroom Deep Clean" mentions deep cleaning tiles and grout, which is cleaning and not installation. The site states no tile size, material, room type or surface for any of the three rows.

No. The catalog has no row for laying planks, vinyl or sheet flooring, no row for underlayment or floor preparation, and no row for fitting baseboards. The word flooring in the route name is the search term teams use, not a task the site states, and this page claims none of it. The floor-related rows the catalog does word are cleaning rows, which are not installation.

Teams that need flooring beyond tiling can still ask for it. The route is the same as for every custom collection: you describe the work, we write a spec, and we quote against it before any capture begins. Whether recording for that work is already running is Not stated, so expect a quote rather than a fixed timeline. Wall tiling versus floor tiling is also Not stated for the one tiling row, so the spec settles which surface your episodes show.

Tiling is about where things sit relative to each other: where a tile lands on its bed, how it meets its neighbour, where a cut edge meets a wall. A stereo pair recovers depth from the calibrated frame pair rather than from a standalone depth sensor, so the form of the floor or wall plane and the position of an edge sit in the same frame as the hands and the tile. The site gives no depth accuracy figure for this row and no flatness, level or joint-width label, so what a buyer does with that depth is their pipeline’s question.

A tile edge is a good test of why geometry matters. A flat RGB frame shows the edge as a colour change between a tile and the surface beside it. A stereo pair adds where the two surfaces actually sit in three dimensions. The site does not state a placement accuracy figure and promises none. It also does not say what a lab would measure from the depth stream, so that is settled in the spec.

The catalog lists a head-mounted phone, which records the first-person view of real work on site, and a multiview rig with handheld grippers, which records a staged working set. On the staged set the operator performs the task with handheld grippers, so the demonstration maps onto a robot end effector. A wrist-mounted camera is available alongside the head camera for close-range detail on the piece being placed. The row lists both modes and does not say which applies, so whether a given batch shows a real workplace or a staged set is agreed in the spec.

The site also describes two stereo setups and does not say which one a tiling 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. Intrinsics and extrinsics are re-solved at the start and end of every session, an episode is accepted only if RMS reprojection stays under 0.28 px, and streams are hardware-triggered onto one clock with a 2.0 ms cross-stream ceiling. The second table below sets what is stated beside what is not.

Hand pose is delivered for both hands, as 21 joints per hand, each with a visibility flag. On a tiling task that is the label that carries how the fingers press and adjust a piece, how a spacer is placed, how the other hand steadies a tile or a cutter, and how the grip changes between spreading, setting and grouting. The site’s tool use page says grip changes live in finger-level detail that RGB video alone cannot resolve, which is the reason hand pose sits beside the video.

Contact events mark the moment the hand meets an item. The site does not say whether a tile, a spacer, an adhesive spreader, a cutter or a grout tool counts as the item, so the spec is the place to settle what a contact event means for your task. The row names adhesive, tiles, spacers, edge tiles and joints, and the site names no trowel, float, plank or cutting tool model, so no tool is promised here.

The row words four operations, and they use the hands and the tool differently: spreading adhesive, setting tiles with spacers, cutting edge tiles and grouting the joints. Each depends on the one before it, so a good spec treats each as its own task with a start state, an end state and accept criteria. That keeps action segments consistent across contributors and lets you ask for more of one operation without re-collecting the others.

Stereo helps differently at each step, and the third table below sets that beside what stays open. Episode boundaries are a spec choice too: one tile placement, one row of tiles, one cut, one grouted area or one full run from adhesive to grout. The site states no task duration, cycle time or throughput for tiling, so none is given here.

Wet and reflective surfaces are the first risk. Spread adhesive and grout are wet while they are worked, and the site says depth degrades on wet and reflective surfaces. 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 a particular bed of adhesive or on narrow joints, so this page promises no figure. The statement that painted and oily surfaces absorb a depth projector pattern is made for under-vehicle capture only, and it is not extended to tiled surfaces here.

Hands are the second. A tile, a spacer or a tool sits against the surface and the fingers wrap it, so the contact point is often hidden. The visibility flag on each joint is the answer the site gives, and a spec can ask for wrist views so a grip the head camera loses is still recorded. The piece being placed can also fall out of frame. An episode is rejected when the manipulated object is out of frame for more than 15 percent of it, when sync breaches the ceiling for more than three consecutive frames, when calibration drifts beyond tolerance, or when an undocumented bystander appears, and a reject log ships with every batch.

Calibration is the third, because stereo depth is only as good as the rectification behind it. Floors and walls are large plain planes, and the site states nothing about how that affects depth. The site states no glare threshold and no motion blur threshold for fast hand movements, so neither is promised here. The fourth table below sets each failure beside what the site states about it.

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 tiling 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. The fifth table below sets each stream beside what the site states about it.

Validation runs against signed accept criteria written into the spec before capture. Each batch goes through per-batch QA and a reject log ships with it. Only footage that clears the criteria is billed as a validated hour, and rejected items are recaptured at our cost. A human reviewer checks each episode against the task spec. For a tiling task, the spec decides what the manipulated object is: the tile, the tool or the area being laid.

Every contributor signs a written release before capture, and each delivered episode links to that record. At ingest, automated detection finds faces, plates and screens and blurs them irreversibly, and a human reviewer confirms that no un-consented person is identifiable in any frame. The site does not state that readable signage is blurred, so raise any such need before the spec is signed. Construction sites can have other people nearby, so settle bystander handling in the spec for the setting you choose. The site states no contributor safety protocol, no dust or building-code claim and no safety certification for tiling, and this page describes none.

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 tiling. 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. The tiling row is Active collection, which means recording is running, but the site does not state how deep coverage is, and the construction guide treats construction and skilled trades as a thinner-coverage setting, so plan on a quote rather than assuming the faster figure. A pilot reaches its first delivery in two to three weeks and a standing program reaches full rate in four to six weeks, assuming capture coverage is in place. Collections are billed per validated hour, and capture at your own facility is listed as not offered. After the first batch, on-demand top-ups run against the same accept criteria.

Describe the tiling operations and their order, what counts as a completed placement, the surface and the setting, and the variation your pipeline is missing, such as edge cuts or grouting. 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, zarr or Rerun .rrd.

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

Where this site does not state something, this page says Not stated rather than filling the gap. There is no listed label for tile size or material, adhesive or grout product, cure time, joint width, flatness, level or placement accuracy, no plank, vinyl, underlayment or baseboard task, and no tile, spacer or tool specification. There is no building-code, dust, ventilation, PPE or safety guidance, and nothing here claims a training outcome or a model improvement.

Force or tactile readings are not in standard capture. The site separately offers a tactile glove tier as a pilot-ready option, with glove signals time-synced to a head-mounted stereo camera, and does not state that it has been used on tile work, so agree the fit first. The tradeoff to plan for is the embodiment gap between a human hand and a robot platform laying tile, which retargeting or pairing with robot data has to close. Counts of contributors, hours and tasks are set in your skill spec.