Egocentric Floor Cleaning and Mopping Video Data

Egocentric floor-cleaning video data is first-person video of a person mopping, sweeping or vacuuming a floor, recorded from the contributor’s own point of view with synchronized sensor streams alongside. The camera sits on the head, close to what a robot’s onboard camera would see while it works a floor, so the frames show the tool, the floor in front of it, the hands on the handle and the obstacles the person steps around, rather than a distant view of someone cleaning a room. On this site it is a custom collection: you name the task, the tool, the floor and the setting, and capture is run to a written skill spec.

The data is human capture. A paid, consented contributor wears the rig and does the task. It is not robot or teleoperation data, and it is not a fixed corpus you download. This page is a different subject from wiping and surface cleaning data, which centres on counters, tables and other surfaces worked by hand, and from household robot pretraining data, which spans whole homes. The subject here is the floor and the tools that work it: mop, broom, vacuum and floor machine. It covers what the site states about capturing, labeling and checking those tasks.

People reach this topic through close cousins: floor cleaning robot training data, a mopping video dataset, egocentric sweeping and vacuuming data, or custom floor-cleaning data collection. They describe one request from different angles, and the work is the same: write down the task, collect it on demand, and receive validated episodes. Nothing on this page states that the video carries a coverage map, a dirt label or a floor-type label.

A published robot foundation model paper, GR00T N1, linked in the sources below, describes a data pyramid with web data and human video at the base, synthetic data in the middle and real robot trajectories at the top. First-person human video of floor work sits in that base layer. Human capture does not replace robot data, and this page does not claim that it does.

A floor task is a long walk with a tool. The person moves through a room or a corridor, the tool follows, and the floor changes as the work goes: wet where it was mopped, clear where it was swept. The site describes household collection as longer-horizon and lower-precision than a bench task, with more whole-home navigation between rooms than any other setting, and it describes the vacuuming row as whole-room navigation with dynamic obstacle avoidance. A clip that shows a hand somewhere near a mop does not carry what a policy needs. The hands, the tool and the movement have to be present in the episode, and the episode has to say where one step ends and the next begins.

Breadth matters as much as volume. A published model write-up, GEN-0, also linked in the sources, puts it this way: "data quality and diversity matters more than sheer volume." For floors that points at the spread of tools, rooms and obstacles. A custom collection lets you aim at the step your model is missing, such as wringing the mop and changing the water, or moving chairs before a vacuum pass, rather than hoping a broad household collection happens to contain it. This page does not claim that any public corpus lacks floor work.

The first table below reproduces fourteen rows of the site’s commercial task catalog that name mopping, sweeping, vacuuming or floor care. Each is listed with a skill level, a capture setup and a status exactly as the catalog states them. It is a selection, not the whole catalog. Vehicle interior vacuuming rows are left out because a cabin is not a floor, and the restaurant kitchen deep-clean row is left out because floors are one item in a longer list of surfaces.

Mopping appears in the commercial cleaning and janitorial environment as "Mopping", which fills the bucket, wrings the mop and mops hard floors in sections, changing water as needed. It is easy and marked active collection. The same environment lists "Vacuuming", which vacuums carpets and hard floors in offices and corridors, moving chairs as needed, also easy and active collection. The gym and fitness studio environment lists "Floor cleaning", which mops and disinfects gym floors and mats and wipes down equipment, easy and available on request. Hotel and lodging lists "Lobby & Common-Area Cleaning", which vacuums and mops lobby floors, dusts furniture, polishes glass and empties bins, easy and ready to deploy.

Sweeping appears in several settings. The event venue environment lists two rows, one that sweeps the venue floor, bags trash and recycling and carries the bags to the dumpsters, and one that walks the venue collecting trash and recycling into bags and sweeping up debris. Both are easy and ready to deploy. The hospitality service set lists "Sweep a floor area", which sweeps and collects debris, marked available on request with a multiview rig and handheld grippers. The active construction site lists a row that sweeps up debris, loads it into wheelbarrows and hauls it to the site skip, marked active collection. The hair salon environment lists "Workspace sanitation", which sweeps hair, disinfects tools and chairs and wipes down the station between clients.

Floor machines and carpet care are the medium-skill rows. "Floor Stripping & Waxing" strips old wax with a floor machine, rinses and applies new wax in even coats, and a second row strips and waxes hard floors with a floor buffer and wax applicator. A third extracts and dries carpets with a carpet extractor and air movers. These are catalog listings, and the page does not claim more about them than the catalog does.

The second table shows where the site names floor work in homes and elsewhere. The household chores dataset lists "Vacuum with obstacle negotiation" among its skills, along with surface wipe and clutter reset, but it does not list mopping or sweeping, so a request for a domestic mop or broom is a skill-spec question rather than a stated capability. The site does not describe a floor type, a tool model, a cleaning product or a room size for any row, so none is stated here. A task outside these rows is settled in a Spec Pilot rather than assumed.

Each attempt is delivered as one episode, and a floor task needs a decision up front about what one attempt is: one corridor, one room, one bucket-fill-and-mop cycle, or one sweep-and-bag pass. The site does not set that length, so the skill spec does. Inside the episode, the synchronized streams run from start to end, and action segments divide it into labeled spans.

A mopping episode might open with filling the bucket and close with the last section mopped, and a vacuuming episode might span a room with a chair moved partway through, as the catalog rows describe. These are illustrations of how a spec can cut an episode, not stated durations. The site publishes sample footage of an operator carrying a cordless vacuum down a staircase into a living room, listed with the household chores dataset, which shows the kind of movement between spaces that a vacuuming episode can include.

The household chores dataset names its environments as living space, bedroom, bathroom, utility room, and stairs and hallway. The commercial rows name offices and corridors, lobbies, gyms, event venues and guest rooms. Which of those apply to your program is set in the spec, along with how many rooms, floors and tools you want covered.

A floor-cleaning collection uses the same capture stack as every Firsthand program. The instrumented tier is a head-mounted stereo rig that records first-person RGB, with a metric depth stream recovered from the calibrated stereo rig rather than from a standalone depth sensor, and 6-DoF IMU data at 200 Hz from head- and wrist-mounted sensors. The wider stack is ten streams: a 4K head camera, two wrist cameras, depth, IMU, 3D hand and body pose, instance segmentation and action labels, hardware-aligned under 2 ms. The two wrist cameras give a closer view of the hands than the head camera alone, which matters where a hand grips a handle or moves a chair.

Every stream is hardware-triggered onto one PTP clock domain, so depth, motion and hand pose line up with the same video frame rather than being approximated afterwards. The rig is calibrated twice per session, before capture starts and after it ends, and an episode is accepted only if RMS reprojection stays under 0.28 px. The catalog lists the capture setup per task: a head-mounted phone or multiview rig for the cleaning, hotel, venue, construction and salon rows, and a multiview rig with handheld grippers for the hospitality service set row. Which tier applies to your program is set in the skill spec.

The site does not state a camera count, resolution or frame rate specific to floor work. Force and tactile readings are not among the streams, no mapping or localization stream is among the ten, and no robot or teleoperation data is captured.

Inside each episode, action segments divide the recording into start- and end-timestamped spans, each carrying start_ns, end_ns, verb and noun. For this task family a segment can be one step as written in the skill spec, so the mopping row becomes a run of labeled steps such as fill, wring and mop rather than one long clip, and a sweeping row can separate sweeping from bagging. The other standard fields are hand_joints and hand_visible for 3D hand pose, contact_events for the moment the hand meets an object, objects, and instance_masks, stored at the depth resolution of 848×480 so they line up with the depth stream.

Hand pose is delivered for both hands with a per-joint visibility flag for occluded fingers. That is the label that carries how the hands sit on a handle and where they move during a pass. The site describes contact events as the moment the hand meets an item, and it does not say whether a handle, a bucket, a chair or the floor counts as the item, so the spec is the place to settle what a contact event means for your task.

The site does not list labels for the floor itself. There is no stated label for a floor type, a cleaned-area or coverage map, a dirt, stain or wet-patch annotation, a contact force or pressure reading, or a task-completion score. Catalog descriptions such as mopping hard floors in sections or moving chairs as needed describe the task the contributor performs, not a delivered measurement.

Several limits the site already states apply directly to floor capture, and several it leaves open. Each stated limit below appears elsewhere on the site rather than being introduced here.

Hand occlusion is the first. Fingers disappear behind objects in manipulation, which is why 3D hand pose carries a per-joint visibility flag. The second is depth, which the site says degrades on wet and reflective surfaces and handles with per-frame confidence maps. A freshly mopped floor is a wet surface, but the site does not say how depth behaves on a wet floor in particular, so this page does not promise a figure for it.

The third is the set of machine-checked reject criteria. An episode is rejected for a sync breach beyond the ceiling for more than three consecutive frames, calibration drift beyond tolerance, the manipulated object out of frame for more than 15 percent of the episode, more than 35 percent invalid depth pixels in the manipulation region, an undocumented bystander, or a mismatch against the named skill spec. A human reviewer then checks each episode against the task spec. For a floor task, the spec decides what the manipulated object is, the tool or the patch of floor being worked.

What the site does not state is just as useful to know before you scope: no mask accuracy figure for floors, tools or wet patches, no description of how a wet-floor warning sign or a floor machine is handled, and no cleaned-area annotation. If your policy depends on any of those, say so at the start.

Validation runs against signed accept criteria written into the spec before capture. Each batch goes through per-batch QA and a reject log ships with it. Only footage that clears the criteria is billed as a validated hour, and rejected items are recaptured at our cost. Many teams start with a Spec Pilot, a small, fast first batch that proves the spec and the accept criteria on real footage before volume scales.

Every contributor signs a written release before capture, and each delivered episode links to that record through consent artefact IDs in the batch data card. The consent chain has six checkpoints, from before capture to after delivery, and a break at any one discards the footage rather than shipping it at a lower grade. At ingest, automated detection finds faces, plates and screens and blurs them irreversibly, and speech is scrubbed of names, addresses and card numbers while ambient audio is kept. A human reviewer confirms that no un-consented person is identifiable in any frame, and footage that cannot be de-identified is discarded. The anonymization record is delivered with the data.

Floors run through shared spaces, so bystanders matter. The site states that household capture happens inside lived-in spaces and that bystanders are de-identified or the footage is discarded. For the household dataset it adds that bathroom capture excludes any episode with a person in frame other than the consenting operator. Hotel lobbies, corridors, gyms and event venues are public-facing settings in the catalog, and the hospitality page describes hospitality and commercial housekeeping as one of the thinner-coverage environments, so settle bystander handling in the spec.

On safety and qualification, the site states no contributor safety protocol, no guidance on wet-floor hazards, cleaning products or floor machines, and no safety certification for any floor row, and this page does not describe one. The site lists capture at your own facility as not offered. If your program needs terms about who performs the work, the equipment used or the conditions under which it is done, raise them when you scope it. None of this is legal advice, and the executed agreement controls.

From signed spec to first validated batch, the timeline is a median of 48 hours where coverage is deep and three to four weeks where it is thin and contributors need to be recruited first. Speed comes from a warm pipeline, not from cutting consent, review or anonymization. Because the hotel and hospitality rows sit in an environment the site calls thinner-coverage, ask which timeline applies to the setting you want when you request the quote. After the first batch, on-demand top-ups run against the same accept criteria, so you can add more rooms, tools or floors without rewriting the spec. The site does not state a turnaround for those later batches, so plan them in your quote.

Delivery is in the format your stack already reads: RLDS, LeRobot, WebDataset, HDF5, zarr or Rerun .rrd, or raw streams with metadata. Readable converter source ships alongside the data so the schema can be retargeted to your own loader. Collections are billed per validated hour.

The license is perpetual, irrevocable and worldwide, with the data buyer-owned and non-exclusive by default. Training, evaluating, fine-tuning and deploying models, including commercial ones, is permitted, and exclusivity is available as a separate term. The executed agreement controls. The site does not publish a price list, a minimum order or a batch size for this task family, so those are set in your quote.

Where this site does not state something, this page says "not stated" rather than filling the gap. The table below separates what is delivered from what is not stated. There is no listed label for floor type, coverage, dirt or wet patches, no force or tactile stream, no tool, product or floor specification, and no room or floor count. Nothing here claims a training outcome or a model improvement.

The tradeoff to plan for is the embodiment gap between a human holding a handle and a robot platform working a floor, which retargeting or pairing with robot data has to close. Counts of homes, contributors, hours and tasks are set in your skill spec.

Describe the tools, the floors, the setting and the variation or failure cases your pipeline is missing. That becomes a written skill spec, quoted for reach, timeline and price before any capture begins, and most programs run a Spec Pilot first. The scoping checklist below maps the parts of a skill spec onto a floor request. A published sample recording is on the sample data outputs page if you want to see the streams before you write a spec.

Three choices matter most for this family. Name the step and the tool rather than a broad activity, because that is what lets each step become one action segment. Name the setting and the capture tier, since a hotel lobby, a gym, an event venue and a living space present different geometry and the catalog lists a different capture setup for some of them. And name whether the task must be a home task: the household dataset lists vacuuming but not mopping or sweeping, so a domestic mop or broom request is settled in the spec.