Egocentric Vehicle Maintenance Video Data

It is first-person video of a person servicing a vehicle: draining and refilling oil, swapping a filter or battery, replacing brake pads, or connecting a scan tool and reading fault codes. The camera is head-mounted, with two wrist cameras for a closer view of the hands. The contributor is a paid, consented person following a written skill spec, not an actor and not a robot.

Where the site states nothing, a table below says "Not stated".

The task catalog already names maintenance work. Thirteen rows sit in one setting, Auto Service Center & Tire Shop, in the Automotive & Transport sector, whose examples read "Car garage, dealership service bay, oil change, tire rotation, brake work, vehicle diagnostics." The table below reproduces the rows as the catalog states them.

They fall into five groups: three oil and fluid rows, two filter rows, three battery rows, three brake rows, and two diagnostics and general rows. One is listed as easy, seven as medium and five as hard. All use the capture setup "Phone (head-mounted) / multiview rig".

Four rows give a setting, "Automotive repair shop with vehicle lifts, tool chests, and dedicated service bays", where the other rows give steps. They are reproduced as listed; the site says nothing more about how lifts are used. Three brake rows also mention removing a wheel, so they overlap the tire and wheel service page.

Tire and wheel rows and washing and detailing rows sit in the same catalog but have their own pages, so they are left out. The catalog lists no row for replacing wipers or bulbs and none for a multi-point inspection.

An episode is one contributor performing one task from the skill spec, such as one oil change or one battery swap. The wider stack is ten streams, including a head-mounted global-shutter 4K/60 camera with a 120 degree lens, two wrist-mounted 1080p/60 cameras, depth, 200 Hz 6-DoF IMUs, 3D hand and body pose, instance segmentation and action labels, hardware-aligned within the 2 ms ceiling the site documents.

Every episode carries int64 nanosecond timestamps per stream, camera intrinsics and extrinsics, hand joints for both hands, instance masks, action segments and contact events. The rig is calibrated at the start and end of every session, and an episode is accepted only if RMS reprojection stays under 0.28 px.

The unit of labeling is the action segment: a taxonomy of 97 verbs and 512 nouns, annotated start and end to the frame and human-reviewed. A spec that says "change the oil" gives one long segment. A spec that names drain, swap the filter and refill gives a sequence you can slice by step.

Data is delivered in RLDS, LeRobot, WebDataset, HDF5, zarr or Rerun .rrd, with converters shipped as readable source. You receive a perpetual, irrevocable, buyer-owned commercial license, the consent artefacts for the batch and a data card that lists jurisdictions of capture. Billing is per validated hour, and rejected episodes and recaptures are never billed.

The site states no maintenance learning result, so this is plain reasoning about the task, not a finding. A job is a chain of dependent steps, and the order matters: the filter comes off after the drain, and the terminals are tightened after the swap.

Two consequences show up in the data. Sequence: action segments keep each step in order and sliceable. Occlusion: the site’s automotive service page says the vehicle itself frequently occludes the view, and that under-vehicle work happens in tight, low-light cavities where depth degrades, so those episodes are flagged for depth quality instead of silently interpolated.

Firsthand captures people, not robots. There is no robot joint-angle log, robot result, benchmark or teleoperation data, and using the episodes for a specific robot means your own retargeting step. The standard capture lists no force or tactile stream, only contact events. A separate glove tier has its own page, and this page makes no claim about it for maintenance work.

The catalog places every maintenance row in Auto Service Center & Tire Shop. The site’s automotive service page lists bay setups as lift bay, tyre bay, engine bay work and under-vehicle on lift. It does not state which sites or countries host capture, or how many, and it states no capture at a buyer’s own garage or facility. The spec settles location and access.

Only what the site already states applies to privacy. Every participant signs a written release granting commercial training rights before capture begins. Faces and license plates are detected and irreversibly blurred, and the automotive service page adds VIN plates, blurred at ingest without exception. A human reviewer confirms that no un-consented person is identifiable in any frame, audio is scrubbed of identifying speech, and a clip where a bystander cannot be de-identified is discarded. The site says nothing separate about customers or vehicle owners. This is not legal advice, and the executed agreement controls.

The site states one published timeline: from a signed spec to a first validated batch, a median of 48 hours where coverage is already deep, and typically three to four weeks where coverage is thin because operator vetting is the bottleneck.

Four of the thirteen rows are marked ready to deploy, meaning contributors and site access are set up and capture starts once a spec is signed. The other nine are marked active collection, meaning recording is running now, so a spec can join an existing capture stream. The site does not say which timeline applies to which row, so ask when you request the quote.

Start with the free, ungated sample pack to check the loader and the label quality, then move to a Spec Pilot: a small first batch against the written spec that proves the fill rate and the reject rate on your conditions before you scale.

A good brief names the maintenance task, the order of steps, how each step starts and ends, and what counts as a reject. The site’s skill-spec guide says to name the verb and the noun rather than a broad activity. Then request a custom dataset.