Stereo Grinding and Deburring Data

It is first-person video of a person smoothing metal after it has been cut or welded: grinding weld seams flush with an angle grinder, then finishing edges and removing burrs with files. It is recorded from their own point of view with two calibrated cameras, with hand pose and sensor streams alongside. The task is the one the site’s catalog lists in two rows. Stereo describes the capture, and MCAP describes the container. They are separate choices.

The site does not state that stereo is already running on these rows. 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. It describes task video only and makes no surface-finish, tolerance, safety or certification claim. Firsthand is not affiliated with Foxglove or the MCAP maintainers.

Two, both in the Metal Fabrication & Welding Workshop environment of the Manufacturing & Workshops sector, both rated Medium and both listed as Ready to deploy. The first is named "Grind and deburr welded seams using an angle grinder and files": "Grind weld seams flush with an angle grinder and finish edges with files." The second is named "Grinding & Deburring": "Grind welds and cut edges smooth and remove burrs with grinder and hand files."

Both list a head-mounted phone or a multiview rig as the capture setup. The catalog lists the environment examples as a metal shop with welding bays, grinders, cutting and fitting benches, and does not say which shop a row was captured in. The site says nothing further about what Ready to deploy means in volume, hours or delivered stereo footage, so this page claims none. It names only the tools and items that appear in the rows: an angle grinder, files, welded seams, welds, cut edges and burrs. Other rows mention grinding as one step in a longer task, for example weld inspection and cleanup, and are not counted here. The first table below reproduces the two rows as the catalog states them.

The stereo welding and metal fabrication page gives stereo and MCAP treatment to twelve rows in five families, and the two grinding rows are one of those families. The stereo surface finishing page covers sanding, drywall and painting, and leaves deburring rows out of its count.

This page is the stereo and MCAP treatment of the two grinding and deburring rows alone, with their own stream map and failure modes. The comparison table below sets the pages side by side.

Grinding is about where the tool meets the part: how far a seam stands proud, which edge has been finished, where a file has been. A stereo pair recovers depth from the calibrated frame pair rather than from a standalone depth sensor, so the seam, the tool and the hands share a frame. The site gives no depth accuracy figure for these rows, no surface-roughness or tolerance figure and no finish-quality label, so what a buyer does with that depth is their pipeline’s question.

The site also states that depth degrades on wet and reflective surfaces and handles that with per-frame confidence maps shipped so a buyer can mask rather than guess. How depth behaves on bare, polished or ground metal is Not stated, and a spec should check it on a Spec Pilot. Capture at your own facility is listed as not offered.

The site states 3D hand pose of 21 joints per hand with a per-joint visibility flag, and contact and grasp events that mark when fingers meet a tool or a part. In grinding and deburring that is the two-handed guide of an angle grinder, the hold on a file and the fingers steadying a workpiece.

The site does not say whether a grinder, a file or a workpiece counts as the object for contact events, so settle it in the spec. Wrist cameras give a closer view of the hands, and a multiview rig adds synchronized third-person angles when the spec asks for them.

The site describes two stereo setups and does not say which one a grinding 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 grinding order uses is Not stated on the site, and the third 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. The site does not describe how a recording handles sparks, dust or noise, and gives no heat, spark or protection rating for any rig, so any requirement is settled in the spec.

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 grinding work the site states hand pose, a stereo-derived depth stream with confidence maps, 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 fourth table below sets each stream beside what the site states about it.

Reflective surfaces are the first risk. The site states that depth degrades on wet and reflective surfaces, with per-frame confidence maps to mask low-confidence regions, and an episode is rejected when more than 35 percent of depth pixels in the manipulation region are invalid. How a stereo pair behaves on bare, polished or freshly ground metal is Not stated, and this page makes no claim about it.

Occlusion and blur are the second. A grinder guard or a clamped workpiece can hide the fingers that hold a tool. The visibility flag on each joint is the answer the site gives, contact events mark when a hand takes hold of an item, and a spec can ask for wrist views or a second angle so a grip the head camera loses is still recorded. The head camera is global-shutter, which avoids rolling-shutter skew, and the site notes that reach and grasp frames are where blur and occlusion show up.

Sequence is the third. The rows chain grinding a seam flush and finishing edges with files, so a spec naming each step gives a sequence that can be sliced by step, while a spec saying "grind and deburr the part" gives one long action segment. The rows do not say what one episode covers, so the spec decides. Grit sizes, disc types, speeds, spark, dust, noise and PPE handling, safety certification, surface-roughness or tolerance figures, shop or customer names and throughput are Not stated and are not claimed on this page.

Calibration is the fourth, 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 when an undocumented bystander appears, and a reject log ships with every batch.

For every delivered episode the site states that faces, plates and screens are blurred irreversibly, and that footage that cannot be de-identified is discarded. The anonymization record ships with the data, and an undocumented bystander is a stated reject criterion.

The site does not state a rule for shop access or work areas, so this page makes no such claim. Raise it in the spec if it matters.

The October 2026 off-the-shelf catalog lists ~324,800 hours across 18 datasets, of which stereo is ~64,800 hours across 9 datasets. The site lists no off-the-shelf grinding or deburring dataset, so stereo grinding and deburring in MCAP is custom collection: custom collection at scale through a network of thousands of capture providers, for a consistent, recurring data flow to labs. The off-the-shelf stereo and hand-tracking video page shows what the rig delivers, and the custom order applies it to your task.

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. Both grinding rows are Ready to deploy, and the site does not say which speed applies to a stereo grinding order, so the timeline is quoted against your spec. Collections are billed per validated hour, and rejected episodes and recaptures are never billed.

Describe the steps and where each starts and ends: grinding the seam flush, then finishing edges and removing burrs with files. Name the tools you expect to see, whether an episode covers one seam or a whole part, what counts as a reject 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 or WebDataset.

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 fifth table below shows what each spec element decides.