Egocentric Knot Tying and Rope Handling Data

It is first-person video of a person handling something long and flexible: a lace through an eyelet, a cable coiled and strapped, a bundle dressed with ties. 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.

In robotics a rope, cord, lace or cable is often called a deformable linear object: much longer than it is wide, with no fixed shape, so its form changes as the hands move it. This page uses the term only in that plain sense. The value of the data is what the hands do to the cord, and it is labeled so each step can be sliced out. Where the site states nothing, a table below says "Not stated".

The task catalog is the starting point, and it does not name a knot or a rope. No row describes tying a knot, tying a shoelace, coiling a rope or securing a load with a strap. The rows that do involve cord-like items are these.

Three sit in the Deformables & Textiles Bench, in the Manufacturing & Workshops sector, whose examples read "Cloth-and-cable bench: garments, towels, cables, form boards, bags." "Coil + Velcro cable" is described as "Over-under coil and strap a cable" and is listed as medium skill. "Thread lace eyelets" is described as "Thread a lace through eyelets" and is listed as hard. "Route + clip cables" is described as "Route cables to a form board" and is listed as medium. All three use the capture setup "Multiview rig + handheld grippers" and are marked available on request.

A fourth row, "Cable-comb dressing" in the Data Center & IT Hardware Bench, is described as "Dress a cable bundle to uniform pitch with ties" and is listed as easy, with the same capture setup and status. Separately, the household task list names "bin change and bag tie" next to laundry and bed-making, with no description beyond that name, so the site does not say whether a bag is tied with a knot, a twist or a drawstring.

The bench lists 12 rows in all. The others cover buttons, a zipper, folding, flattening, hanging, socks, a duvet cover and wrapping, which are fasteners and fabric rather than cord, so they are left out. Hose rows, such as washing dishware with a spray hose, are cleaning tasks and are left out too. Anything beyond the five tasks above, including a knot, a rope, a strap around a load or a particular lace pattern, is a skill-spec question to settle in a Spec Pilot rather than an assumption.

An episode is one contributor performing one task from the skill spec. 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 site does not state a cord path, a rope shape or a knot state, or how thin a cord can be before depth becomes unreliable. It does say that metric depth on sub-centimetre parts is flagged rather than interpolated, but that statement is about small fasteners and parts, so this page does not extend it to cords.

The unit of labeling is the action segment. The site states a taxonomy of 97 verbs and 512 nouns, annotated start and end to the frame and human-reviewed. A spec that says "thread the lace" gives you one long segment. A spec that names lay out, pass through each eyelet, pull and finish gives you a sequence you can slice by step.

Hand pose carries a per-joint visibility flag, and contact events mark when a hand meets the item. The site states that clothing, bedding and bags do not track well with masks propagated frame to frame, so deformable captures ship per-frame instance masks where the spec calls for them. Rope, cord and lace are not named in that statement, so whether a cord task asks for them is a spec decision.

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 knot-success or quality label.

The site makes the central point in its own words: a cable has no single shape. A lace or a coiled cable shares that, so position and orientation alone do not describe the object. Three consequences show up in the data, and each is something the site already addresses.

First, occlusion: fingers close around a cord and disappear, so the visibility flag marks joints that were not seen instead of presenting a guess as a measurement. Second, coordination: one hand holds while the other works, so both hands need to be in frame at once. Third, tracking: masks are the stated answer for items that do not track well when propagated, and the spec is where you decide whether a cord needs them.

Firsthand captures people, not robots. There is no robot joint-angle log, robot result, benchmark or teleoperation data for these tasks, and using the episodes for a specific robot means your own retargeting step. Nothing here claims a training outcome.

The standard capture lists no force or tactile stream. Contact is described only by contact events, and cord tension is a force quantity for which the site states no label. A separate glove tier is described on its own page, and this page makes no claim about it for cord work.

The catalog places the cord-related rows in a Manufacturing & Workshops bench and a Data Center & IT Hardware Bench. The site does not state which benches, sites or countries host capture, or how many, and it states no capture at a buyer’s own 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 licence plates are detected and irreversibly blurred, 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 rather than shipped. 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.

All four catalog rows above are marked available on request, which the site describes as scoped and recruited to spec when a buyer asks for it. The site does not say which timeline applies to these rows and promises nothing faster, 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 cord 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.