Egocentric Data Center and IT Hardware Task Video Data

Egocentric data center and IT hardware task data is real-world video of work at a server rack and a hardware bench. It is recorded against a written spec, with depth, contact events and frame-accurate action segments aligned on one hardware clock where the spec includes them. The tasks it covers are the ones the site states in its task catalog: twelve rows in a Data Center & IT Hardware Bench environment, from patching network cables to swapping a power supply module.

The cable routing and wiring page covers cable work across several settings, and the electronics and device repair page covers consumer device repair. This page covers rack and bench work only.

The site states rack and bench work in its task catalog under the Office & IT sector. The Data Center & IT Hardware Bench environment lists its examples as "Server rack plus hardware bench: modules, cabling, drivers", and it holds twelve rows.

Every row uses the same capture mode, a multiview rig with handheld grippers. Nine rows are rated Easy and three are rated Medium: clipping cage nuts and a panel, inserting a transceiver and fiber, and reseating an add-in card. All twelve are listed as Available on request.

Four rows cover cables and connectors. The catalog descriptions read "Dress a cable bundle to uniform pitch with ties", "Seat power cables and locking clips", "Seat a transceiver and connect an LC fiber" and "Insert and route Ethernet patch cables cleanly." They are rated Easy, Easy, Medium and Easy.

Cable lengths, cable counts, port counts and the layout of the rack are Not stated, and so is any measure of a clean result.

Five rows cover modules and cards: "Remove and insert a drive caddy in a bay", "Unbracket, reseat, and secure an add-in card", "Insert memory modules into slots until they click and latch", "Release and replace a hot-swap fan" and "Replace a hot-swap power supply." Only the add-in card row is rated Medium; the other four are Easy.

Server models, bay and slot counts, and the order of the steps are Not stated.

Three rows cover mounting and marking: "Clip cage nuts and mount a blanking panel" (Medium), "Fasten rail brackets with a driver" (Easy) and "Label a unit and scan the barcode" (Easy). The mounting rows name a driver, but the site states no tool model, torque or fastener type.

Scanner type and label formats are Not stated.

Routing and clipping cables through harnesses and wiring outlets sit on the cable routing and wiring page, and repairing consumer devices sits on the electronics and device repair page. Board-level assembly sits on the electronics assembly page. This page lists them as neighbours and does not count them as data center rows.

Tasks people often expect here, such as racking a full server into a cabinet, configuring a switch or decommissioning hardware, are not stated as catalog rows.

The catalog describes steps that end in a physical seat: memory inserted "until they click and latch", a drive caddy removed and inserted in a bay, an add-in card unbracketed, reseated and secured. Lighting, rack density, reach and cable congestion are Not stated. What the site does state is what egocentric video puts into the training distribution: hands entering the frame from below, occlusion during grasp and motion blur during reach.

The site also states why a hand pose stream matters. RGB alone leaves a model to infer 3D hand structure it cannot see reliably through occlusion and motion blur. Firsthand delivers 3D hand pose as handJoints[2][21][3] in metric units in the head-camera frame, with a visibility flag per joint, aligned to depth and RGB. Whether that stream is populated for the staged gripper capture mode is Not stated, and is agreed in the spec.

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." The site states a head camera at 3840 × 2160 and 60 fps for its instrumented rig, but no camera count, resolution or frame rate for the rack rows, so the capture tier is agreed in the spec.

The site states no gripper model, jaw type or end-effector mapping for these rows. Whether a head-mounted capture mode can replace the gripper mode for them is Not stated, and is settled in the written spec.

Each episode follows the Firsthand episode schema: ten streams synced under 2 ms, with head and wrist RGB, depth, 3D hand pose with per-joint visibility, contact and grasp events, and frame-accurate action segments. For rack and bench work, contact events mark when a part meets a bay, slot or rail, and action segments mark each step.

The site does not state which of those streams the staged gripper mode populates for each row. Object labels use the terms your spec sets. Force or tactile data, and success or quality labels for a seated card, a latched module or a dressed bundle, are Not stated.

The site offers a tactile glove tier as a pilot-ready option. The glove signal is time-synced to a head-mounted stereo camera by hardware timestamps and an IMU, and each glove is calibrated every session.

The site does not state that the glove tier has been used on any rack row, and every rack row lists handheld grippers as its capture mode. Treat force and tactile data as Not stated for these rows unless a spec adds the glove tier and agrees the fit with you.

The catalog places the rows in a server rack plus hardware bench environment. The site names no data center, colocation facility, customer or operator, and it does not state capture at a customer own site or facility. Environment, access and schedule are agreed in the spec, and the number of sites is Not stated.

Every episode links to a signed contributor release through a documented consent chain. Faces, plates and screens are detected and irreversibly blurred at ingest, a reviewer confirms the result before anything ships, and the anonymization record is delivered with the data. An undocumented bystander is a stated reject criterion. How asset tags, serial numbers, barcodes and cable labels are handled is Not stated, so name them in the spec.

Episodes are machine-checked at ingest against the stated reject criteria, listed in the table below, and a human reviewer then checks the result.

The site states no rule for how often a part or port leaves the view in a rack, and no blur or sharpness rule. A rejected item goes to the reject log and is recaptured at our cost.

Delivery formats are RLDS, LeRobot and WebDataset, plus HDF5, zarr and Rerun .rrd. The format table below sets each against what its official documentation states.

Every batch is validated against signed accept criteria before it ships, with a reject log, and delivery is rolling, per validated batch.

Where coverage is already deep, the first validated batch lands in a median of 48 hours from signed spec. Thin environments typically take three to four weeks, because vetting operators there is the bottleneck, and we quote that up front.

On-demand top-ups run against the same spec and accept criteria after each training run. The site states no capture timeline for the twelve rows beyond a quote against your spec.

You describe the rack and bench tasks, the sequence and the capture mode your policy needs. We turn that into a written spec covering tasks, environments, capture tier, annotation depth and accept criteria, then quote reach, timeline and price before any capture begins.

A Spec Pilot, a small and fast first batch, can prove the spec and accept criteria before you scale. Billing is per validated hour, and every collection ships under a perpetual, buyer-owned commercial license.