Stereo Data Center Rack and Server Hardware Data

It is first-person video of a person doing the rack and bench tasks the catalog lists, recorded from their own point of view with two calibrated cameras and with hand pose and sensor streams alongside. 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: 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 data center and IT hardware page covers the same twelve rows as task video data. It names stereo only for the tactile glove tier and has no MCAP content. The cable routing and wiring page covers cable work across several settings, including three of these bench rows, and describes the head-mounted stereo rig used for cable work, with no MCAP content.

This page is the stereo and MCAP treatment of all twelve rack and bench rows together. It states no server model, rack standard, part number, access rule or certification, because the site states none.

Twelve, all in the Data Center & IT Hardware Bench environment under the Office & IT sector, described as a server rack plus a hardware bench with modules, cabling and drivers. Nine are rated Easy and three are rated Medium: "Clip cage nuts + panel", "Insert SFP+ + fiber" and "Reseat PCIe card". All twelve are Available on request and list a multiview rig plus handheld grippers.

The tasks on this page are the ones the rows name, for example "Hot-swap drive caddy" (remove and insert a drive caddy in a bay), "Insert SFP+ + fiber" (seat a transceiver and connect an LC fiber), "Screw-mount rack rails" (fasten rail brackets with a driver) and "Seat RAM DIMMs" (insert memory modules into slots until they click and latch).

The site says tasks people often expect here, such as racking a full server into a cabinet, configuring a switch or decommissioning hardware, are not catalog rows. Tasks outside the list can be added to a spec, but this page claims no coverage of them.

The rows do not state cable lengths, cable or port counts, the layout of the rack, a server or part model, or any measure of a clean result.

Rack and bench work is about where a small part meets a slot: the angle of a caddy at a bay, the line of a module against a socket, the position of a bracket against a rail. A stereo pair recovers depth from the calibrated frame pair rather than from a standalone depth sensor, so the form of the slot and the part is in the same frame as the hands.

No depth accuracy figure, seating label or pass or fail field is stated for these rows.

The catalog describes the mode as "a staged working set recorded by a multiview rig while the operator performs the task with handheld grippers, so demonstrations map onto a robot end effector." Whether hand pose is recorded while handheld grippers are used is Not stated and is agreed in the spec.

The site describes two stereo setups and does not say which one a rack 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. How they combine with the multiview rig is Not stated.

The first risk is occlusion by the hand itself: a hand covers the slot while the module is pressed home. The site answers with a per-joint visibility flag, contact events, and wrist views where the spec asks for them.

Depth is the second. The site says depth degrades on wet and reflective surfaces, masks low-confidence regions with per-frame confidence maps, and rejects an episode above 35 percent invalid depth pixels in the manipulation region. How depth behaves on rack metalwork, transceivers or fiber connectors is Not stated. Calibration is the third, 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. The site states no handling for glare, low light or motion blur on these rows, so this page promises none.

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. The foxglove/mcap repository is MIT-licensed; that is the MCAP project’s licence, not the licence of a Firsthand delivery.

Firsthand offers MCAP delivery, and dataset 06 is stated as shipping as MCAP. For rack and bench work the site states hand pose, a stereo-derived depth stream, IMU, contact events, action segments, per-frame instance masks and one task instruction per episode 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 here; confirm them in the skill spec on a Spec Pilot batch.

Off-the-shelf covers what is already in stock. The industrial and skilled-trade category lists datasets 02, 03 and 04, covering work such as machinery, auto repair, carpentry, construction, factory floors and inspection, and the stereo category lists the stereo head rig. Neither listing names data center or IT hardware work, so this page claims no off-the-shelf coverage of these 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. The site states no provider count, cadence, hours or volume beyond that wording, so they are quoted against your spec.

Where coverage is deep, the median time from signed spec to first validated batch is a median of 48 hours. Where coverage is thin and contributors need to be recruited first, a first batch typically takes three to four weeks. All twelve rows are Available on request, and the site does not say which timing applies to them, so ask when you request the quote.

For every delivered episode the site states that faces, heads, licence plates and legible screens are detected and irreversibly blurred at ingest, a human reviewer confirms the result, and footage that cannot be de-identified is held or discarded. Every episode links to a signed contributor release through a documented consent chain, and the release covers bystanders. Access terms for a site are set per program.

The buyer owns the dataset under a perpetual, irrevocable, non-exclusive-by-default licence, with exclusivity available. This page makes no electrostatic, safety, data center access or certification claim for any row.

Name the rack and bench tasks, the steps in order, what the camera must see, and the streams, topic naming and encoding your reader expects. Say whether MCAP is the only delivery or sits beside RLDS, LeRobot or WebDataset. Most programs begin with a Spec Pilot, which confirms the spec on real footage.