Egocentric Water Heater Replacement Data

It is first-person video of a person replacing a water heater, recorded from a head-mounted camera, 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.

The point of the data is what the hands do across the whole job: disconnecting, moving a heavy unit into place, connecting and checking. It is labeled so each step can be sliced out and trained on separately or as one long episode. Where the site states nothing, a table says "Not stated" and the answer lives in the spec or the quote.

The task catalog lists one row for it: "Water Heater Replacement", in the Trade Service Calls (Plumbing / HVAC / Electrical) environment. The row is described as "Drain and disconnect the old heater, position the new unit, connect water and power and test." Its skill level is Hard, its capture is a head-mounted phone or multiview rig, and its status is Ready to deploy.

That is everything the catalog states about the task. It names no tool, no heater type, no fuel or power source, no size and no fitting, so the first table reproduces the row as written and the rest of this page says "Not stated" wherever a buyer might expect a detail.

The row gives a natural cut: drain and disconnect the old heater; position the new unit; connect water and power; test. A spec that says "replace the heater" gives you one very long action segment. A spec that names those four stages, and splits the connection into water and power, gives you a sequence you can slice by step.

The step table sets each stage from the row beside the episode field that carries it. The stages are the catalog’s own; how finely a spec splits them, and where each one starts and ends, is a spec decision and not a stated default.

The catalog ties the row to Trade Service Calls, which it describes as a plumber, electrician or HVAC crew on service calls at client sites. The row lists a head-mounted phone or multiview rig. The wider site describes a head camera plus two wrist cameras, a depth sensor and an IMU, shown in the capture table below.

The site does not state which homes or sites host a given row, how access is arranged, or how many contributors are available for it. The spec settles location and access, and the quote settles reach.

The catalog marks the row Hard, the same level as the plumbing fixture, HVAC installation and wiring rows in the same environment. It does not say why a row is Hard, so this page does not explain it beyond what the row itself lists: several linked steps, water and power connections, and a test at the end.

Ready to deploy is the catalog’s status for the row. The catalog does not define the statuses further, so ask what it means for your region and your timeline when you request the quote.

An episode is one contributor performing one task from the skill spec, recorded with ten hardware-synchronized streams that line up within 2 ms. The site documents a canonical episode schema, and every delivery format is a view of it.

Hand pose is stored as 21 joints per hand in metres in the head frame, with a visibility flag for each joint. Contact events record when a hand meets an object. The site does not say whether the heater, a hose, a connector or a valve counts as the object, so settle it in the spec.

Action segments carry a start, an end, a verb and a noun from a taxonomy the site states as 97 verbs and 512 nouns, human-reviewed. Every episode links to its contributor’s signed release by a consent artefact ID, and each batch ships with a data card.

The unit of labeling is the action segment: one start, one end, one verb, one noun. Instance masks track object IDs per frame at 848×480, which helps keep the old unit, the new unit and loose parts apart across a long job.

Every episode is machine-checked, then reviewed by a human against its skill spec, and rejected episodes are recaptured at our cost. The site states no label for heater type, connection type, leak result, test outcome or task success, and no force, torque, temperature, pressure or tactile reading, so ask for any such label at the start.

The site describes a small set of failure modes and what it does about each. Hidden hands: a joint hidden behind the unit is flagged by its visibility bit, not guessed. Untrusted depth: depth ships with a per-pixel confidence map, so a buyer can see where depth near a connection should not be trusted.

Drifting calibration: the rig is re-solved against a checkerboard at the start and end of every session and accepted only under the 0.28 px RMS reprojection gate. Weak episodes: anything that fails the machine check or the human review against the spec is rejected and recaptured at our cost. The site states no further failure modes for this task, such as lighting in a utility space, so those are raised in the spec.

Only what the site already states applies. Every contributor signs a written release, and each episode links to it. Faces and screens are detected and blurred irreversibly, a human reviewer confirms that no un-consented person is identifiable, and footage with bystanders who cannot be de-identified is discarded rather than shipped.

The site does not state how a household or the people in it are handled beyond that general rule, so settle it in the spec before capture. This is not legal advice, and the executed agreement controls.

The egocentric plumbing and pipe fitting page covers ten plumbing and pipe rows, of which water heater replacement is one, along with fixtures, drains and pipe cutting. The egocentric HVAC service page covers the HVAC rows and does not list this one. This page takes the single water heater row and goes deeper on its steps, labels and capture.

The Trade Service Calls environment holds eleven rows, including drain snaking, fixture installation, HVAC installation, thermostat wiring and outlet wiring. A buyer who wants a mix of service calls can combine them in one spec.

Data is delivered in RLDS, LeRobot, WebDataset, HDF5, zarr or Rerun .rrd, or as raw streams with metadata, with the episode schema as the canonical layout. QA runs once, against the canonical episode, before any format is cut. The formats table sets the site’s delivery next to each maintainer’s own documentation, linked under Sources.

The license is perpetual, irrevocable and worldwide, buyer-owned and non-exclusive by default, with exclusivity available as a separate term and the executed order form governing.

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 it is thin because operators have to be staffed. The site does not say which applies to this row and promises nothing faster, so ask when you request the quote.

Collection is on demand, so a focused brief is a normal starting point. One full replacement from drain to test is a reasonable first unit, and a Spec Pilot is a fast way to prove the fill rate and reject rate before you scale.

Firsthand captures humans, not robots. Episodes carry human hand pose and human action labels, so using them for a specific robot means your own retargeting step. The site does not state robot actions, teleoperation data or any claim about a particular model or embodiment for this task.

What the data gives a lab is the human side of a long job: the order of steps, how the hands handle a heavy unit and its connections, and where a step begins and ends. A lab can use that for imitation, for segmenting long tasks or for evaluation. Each of those is the buyer’s own use of the data.

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

A good water heater brief names the setting, the stages of the job, how each stage starts and ends, what counts as a reject and the privacy rule for the space. The site’s skill-spec guide walks through that structure. Then request a custom dataset.