Egocentric Furniture Assembly Video Data

It is first-person video of a person putting furniture together, recorded from the contributor’s own point of view with synchronized sensor streams alongside. The frame shows what the person sees: the parts on the bench, the panel in one hand, the driver or drill in the other and the joint being closed.

The key word is assembly: turning components into a finished unit. In the catalog that means joining frames with glue, clamps and screws, assembling commercial furniture units from components, and assembling a flat-pack panel unit with cam locks.

Firsthand captures humans, not robots. Episodes carry human hand pose and human action labels. The site states no robot actions, robot joint data or teleoperation data for this task family, so using the episodes for a specific robot still needs your own retargeting step.

The task catalog names three rows that are explicitly about assembling furniture. Two sit in the carpentry and furniture workshop, a joinery setting with saws, sanders and finishing stations: "Assemble and join furniture frames using wood glue, clamps, and a drill" and "Commercial Furniture Assembly". Both are medium skill level, on a head-mounted phone or multiview rig, with the status Active collection.

The third, "Assemble flat-pack", sits in the trades workshop bench, a bench with hand and power tools: wrenches, drivers, drill, saw, clamps and fasteners. It is described as "Assemble a panel unit with cam locks", at medium skill level, on a multiview rig with handheld grippers, with the status Available on request.

Active collection and Available on request are different statuses and the catalog does not define either further, so ask which applies to your row when you request the quote. The second table lists neighbouring rows, such as fitting cabinet hardware, that are not assembly rows themselves.

The catalog ties furniture assembly to two ordinary workshop settings, not a showroom or a customer building, and the site does not state capture at a customer’s own facility. Capture is run to your skill spec by paid, consented contributors performing the real task while wearing the rig.

The site does not say which rig produces which streams in a given episode, and it does not state furniture types, brands, part counts or the tools used for any row, so those details are agreed in the spec. The capture setup table lists the hardware as the site states it.

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. The fields that matter most for assembly are listed in the episode fields table.

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 a panel, a cam lock or a fastener counts as the object, so settle what a contact event means for your task 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. For furniture, a segment can be one step as written in the spec. 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, which is why the way you write the spec decides how useful the data is. A spec that says "assemble the unit" gives you one long segment. A spec that names the steps gives you a sequence you can train on and slice by step.

Every episode is machine-checked, then reviewed by a human against the skill spec it was captured for. Rejected episodes are recaptured at our cost, and the reasons ship with the delivery.

The site states no label for furniture type, part count, fastener type, joint fit, assembly completion or task success, and no force, torque or tactile reading. If your objective needs a success or defect label, say so at the start so it can be scoped in the spec.

Fingers disappear behind panels and tools, which is why 3D hand pose carries a per-joint visibility flag. Small fasteners can fall below a depth sensor’s reliable floor: hand pose stays valid at that range, and metric depth on sub-centimetre parts is flagged rather than interpolated. Whether a cam lock or a screw head counts as such a part is not stated.

A validated hour is 60 minutes of footage that passed QA against your spec: sync within the 2 ms ceiling, calibration solved and within drift tolerance, coverage cells filled and consent documented. Footage that cannot be de-identified is rejected rather than shipped.

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 the furniture rows and promises nothing faster, so ask when you request the quote.

Collection is on demand, so a focused brief, such as one named unit assembled start to finish, is a normal starting point. Billing is per validated hour, so off-spec footage is not something you pay for.

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, so a gap in the spec shows up on a handful of episodes.

A good furniture brief names the unit and its steps, how each step starts and ends, and what counts as a reject. The site’s skill-spec guide walks through that structure. Then request a custom dataset and we will quote reach, timeline and price against the spec.