Egocentric Vacuuming Data

It is real-world, first-person video of a person vacuuming: carpets and hard floors in offices and corridors, or seats, mats and trim inside a vehicle. It is recorded from the operator viewpoint, with 3D hand pose, depth, contact events and frame-accurate action segments aligned on one hardware clock. The task is the one the site’s task catalog lists in four rows.

The data is human capture by paid, consented contributors, not robot or teleoperation data. This page is about vacuuming robot training data in the sense of what a policy can learn from watching people do the work, and it makes no claim that any robot has been trained on it.

Four, in three environments. "Vacuuming" is in Commercial Cleaning & Janitorial, rated Easy and Active collection. "Car Interior Vacuuming" and "Vacuum, wipe, and condition vehicle interiors using a shop vacuum and detailing brushes." are in Car Wash & Detailing, both rated Easy and Active collection. "Interior Vacuuming" is in Auto Service Center & Tire Shop, rated Easy and Ready to deploy. All four list a head-mounted phone or a multiview rig as the capture setup.

The site says nothing further about what Active collection or Ready to deploy means in volume or hours, so this page claims no completed collection and no delivered hours. It names only the steps and tools that appear in the rows: vacuuming, a shop vacuum, detailing brushes, and vacuuming, wiping and conditioning vehicle interiors. The first table below reproduces the rows as the catalog states them.

The egocentric commercial janitorial page covers the whole Commercial Cleaning & Janitorial environment and lists Vacuuming as one row among its fifteen. The egocentric floor cleaning and mopping page covers sweeping, vacuuming and mopping floors, and the egocentric car washing and detailing page covers eight rows, two of them interior vacuuming. The egocentric carpet extraction and floor stripping page centres the Medium rows and names Vacuuming as a separate Easy row.

This page gives the four vacuuming rows their own treatment: the two settings, how the capture works around a vacuum, a hose and a brush, what each episode carries and where the failure modes are. The comparison table below sets the pages side by side.

The office row is described as vacuuming carpets and hard floors in offices and corridors, moving chairs as needed. That means the data carries the pass over the floor and the moments where the operator moves a chair before continuing, and a spec decides whether moving a chair is part of the episode or cut out.

The vehicle rows name seats, mats and trim, and the Ready to deploy row adds clearing debris from the cabin. One Car Wash & Detailing row names a shop vacuum and detailing brushes and adds wiping and conditioning, so vacuuming there is one step of a longer interior sequence. The site does not say how a vehicle interior is laid out or which seat or panel is cleaned first, so this page does not describe it and the spec decides the scope.

Vacuuming is repeated, close-range work: a tool in one hand, a surface under it and a reach that changes with every pass. A head-mounted camera keeps the whole reach in frame from the viewpoint a manipulation policy will see from, and two wrist-mounted cameras give a closer view of the hands and the tool.

A multiview rig adds synchronized third-person angles when the spec asks for them. That is useful inside a vehicle, where the cabin itself can hide the hands from the head view. Which angles a vacuuming order uses is set in the spec.

The catalog lists one capture setup for all four rows: a head-mounted phone recording the first-person view of real work on site, with a multiview rig adding synchronized third-person angles when the spec asks for them. The wider stack the site documents for hard environments is a global-shutter 4K/60 head camera with a 120° lens, two 1080p/60 wrist cameras and 200 Hz 6-DoF IMUs. Which of those a vacuuming order uses is Not stated, and the third table below sets what is stated beside what is not.

Streams are hardware-triggered onto one clock, synced under 2 ms. Depth is aligned to video, and distance-invalid pixels beyond the sensor reliable range are flagged rather than interpolated. The site does not state which office, shop or vehicle a collection runs in or on, and it does not state capture at a customer’s own facility, so the environment, access and schedule are agreed in the spec.

Each episode follows the same episode schema as every Firsthand custom collection: 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. The only language data is one task instruction per episode.

For vacuuming, hand pose tracks the operator through the grip on the vacuum or brush and the handling of a hose. Contact events mark when a hand takes hold of an item and when it lets go, and action segments mark the start and end of each step. Because the labels share one clock with the video and depth, a step boundary lines up with the exact frame where it happens. The fourth table below lists each item.

Occlusion is the first risk. A hand, the vacuum or the vehicle itself can hide the fingers and the tool, and the cabin can block the first-person view. The per-joint visibility flag is the answer the site gives, contact events mark when a hand meets an item, and a spec can ask for wrist or multiview angles so a grip the head camera loses is still recorded.

Movement is the second. The site’s published household clip shows an operator carrying a cordless vacuum between floors and describes carrying a tool between rooms as a transit segment that staged datasets cut. Whether a vacuuming order keeps or cuts transit between rooms is set in the spec. How depth behaves around carpet, seats, mats and trim is Not stated, so check it on a Spec Pilot.

Scope is the third. The rows differ: one is a corridor and office row, one names vehicle interiors, one adds wiping and conditioning, and none says what one episode covers. Vacuum brands or models, suction or coverage-rate claims, cleanliness or performance claims, service times, throughput and any safety or certification claim are Not stated and are not claimed on this page.

Every batch is validated against signed accept criteria before it ships, with a reject log. The site states that an episode is rejected when the manipulated object is out of frame for more than 15 percent of it, when more than 35 percent of depth pixels in the manipulation region are invalid, or when an undocumented bystander appears.

Billing is per validated hour, and rejected items are recaptured at our cost. The site does not state an accuracy figure for vacuuming step labels, so what counts as a pass for a step boundary is set in the accept criteria.

Every episode links to a signed contributor release through a documented consent chain, and the consent record is part of the delivery. Faces, plates and screens are detected and irreversibly blurred at ingest, and a reviewer confirms the result before anything ships. The site also states that licence plates and VIN plates in service-bay footage are blurred at ingest without exception. The anonymization record is delivered with the data.

Whether office signage, documents or a customer’s belongings are blurred beyond that general rule is Not stated, and this page makes no claim about it. Raise any such rule in the spec.

The October 2026 off-the-shelf catalog lists ~324,800 hours across 18 datasets. The site’s household chores dataset page lists "Vacuum with obstacle negotiation" among its skills in whole homes, but no off-the-shelf dataset names office, corridor or vehicle interior vacuuming. Vacuuming collection to a spec for those settings is therefore custom collection: 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 or throughput for these rows.

The off-the-shelf catalog shows what the capture rigs deliver, and the custom order applies the same approach to your task. Three rows are Active collection and one is Ready to deploy, and the site does not say how many hours exist for any of them.

Delivery formats are RLDS, LeRobot and WebDataset, plus HDF5, zarr and Rerun .rrd. The first three are training-loader formats, HDF5 and zarr carry arrays for custom pipelines, and Rerun .rrd gives a synchronized playback file for review. Converter source is readable, so you can retarget the schema to your own loader, and the episode schema page lists the fields.

Delivery is rolling, per validated batch, and formats are confirmed when the work is scoped.

Where coverage is deep, the median time from signed spec to first validated batch is a median of 48 hours; where it is thin, turnaround runs three to four weeks. All four rows are rated Easy, and the site does not say which speed applies to a vacuuming order, so the timeline is quoted against your spec before capture begins.

On-demand top-ups run against the same spec and accept criteria after each training run, so new vacuuming data stays consistent with what you already hold. The site does not state a batch size or a later-batch turnaround, so those are quoted against your spec.

Describe the setting you want, office and corridor or vehicle interior, and where each pass starts and ends. Name the tools you expect to see, whether moving a chair or clearing debris is part of the episode, what counts as a reject and the variation your pipeline is missing. We turn that into a written spec covering tasks, environments, capture tier, annotation depth and accept criteria, then quote reach, timeline and price against that document before any capture begins.

Most programs begin with a Spec Pilot, a small first batch that proves the spec and the accept criteria on real footage before you scale. The collection runs through the same seven-stage end-to-end custom collection pipeline as every Firsthand program, under a perpetual, buyer-owned commercial license, with exclusivity available. The fifth table below shows what each spec element decides.