Egocentric Ironing and Garment Handling Video Data

Egocentric ironing and garment handling data is first-person video of a person pressing a garment, hanging it, rehanging it or sorting it, 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 over a pressing surface or a rail, so the frames show what the hands and the fabric do rather than a distant view of someone standing near a pile of clothes. On this site it is a custom collection: you name the task, the item, the setting and the steps involved, and capture is run to a written skill spec.

The data is human capture. A paid, consented contributor wears the rig and performs the task. It is not robot or teleoperation data, and it is not a fixed corpus you download. This page is narrower than deformable-object data in general, which covers laundry, bedding and bags, narrower than household robot pretraining data, which spans whole homes, and it is a different subject from laundry folding data, which centres on folding and stacking. The subject here is pressing, hanging, rehanging and sorting garments, and what the site states about capturing, labeling and checking those tasks.

People reach this topic through close cousins: garment handling robot training data, an ironing video dataset, egocentric clothing manipulation data, or custom textile handling 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 cloth-state, wrinkle, grasp-point or force label.

NVIDIA’s GR00T N1 paper 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 garment work sits in that base layer. Human capture does not replace robot data, and this page does not claim that it does.

A garment is one of the least cooperative things a hand can hold. It has no single shape, it drapes and bunches, and part of it is always underneath another part. A clip that shows a hand somewhere near a shirt does not carry what a policy needs. The hand, the fabric and the moment of contact all have to be present in the episode, and the episode has to say where one step ends and the next begins. That is the difference between footage of someone doing laundry and an episode a model can learn a step from.

Google DeepMind’s ALOHA Unleashed paper adds that its shirt-hanging task is long-horizon, involves a deformable object, and needs several dexterous bimanual behaviors. It is a robot teleoperation study, not first-person human video, and Firsthand has no relationship with its authors. It shows why garment data has to cover many starting states and many steps.

Breadth matters as much as volume. Generalist’s GEN-0 write-up makes a point about data in different words: "data quality and diversity matters more than sheer volume." For garments that points at the spread of items, steps and settings. A custom collection lets you aim at the step your model is missing, such as pressing or hanging, rather than hoping a broad laundry collection happens to contain it. This page does not claim that any particular public corpus lacks garment work.

The first table below reproduces twelve rows of the site’s commercial task catalog that name pressing, ironing, hanging, rehanging or sorting garments and linens. 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. Folding rows are left out because they belong to the laundry folding page, and bed-making rows are left out because the item is a bed rather than a garment.

Ironing appears in the catalog as pressing on industrial equipment. Two hotel and lodging rows are titled "Iron and fold linens using an industrial press and folding table" and "Ironing & Folding", which irons linens and uniforms on a press and folds them into stacks. Both are easy and marked ready to deploy. The commercial laundry and linen service environment lists "Pressing", which presses garments and linens on a steam press and hangs or folds them, marked active collection. The textile and apparel production environment lists "Finishing and pressing", which presses seams and finished garments on an industrial iron or press and folds or hangs them, marked available on request. These are catalog listings, and the page does not claim more about them than the catalog does.

Hanging and rehanging appear in three settings. The deformables and textiles bench lists "Hang + space garments". The retail floor set lists hanging garments in size order and re-hanging and tidying a fitting room. The retail store front-of-house environment lists collecting garments from fitting rooms and rehanging them, and sorting garments on the rail by size and style with the hangers spaced evenly. Sorting appears in the commercial laundry row, which sorts incoming laundry by colour, fabric and customer into carts and bins, and in the hotel row for soiled linens.

The second table shows where the site itself names garment work: the household chores dataset, the catalog and the neighbouring pages. It is there so you can see which statements are about homes and which are about other settings. The household chores dataset lists folding and sorting laundry and bed-making, and does not list ironing or hanging among its skills, so a request for a domestic ironing board or a wardrobe is a skill-spec question rather than a stated capability.

The site does not describe a garment type, a fabric, a press or iron model, a hanger type or a garment count for any of these rows, so none is stated here. A task outside these rows is settled in a Spec Pilot rather than assumed. The catalog is a list of what is captured or available, not a ceiling on what can be briefed.

A garment handling 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 second, closer view of the hands than the head camera alone, which matters when a hand smooths a sleeve or slides a hanger onto a rail.

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 multiview rig with handheld grippers for the bench and retail floor rows, and a head-mounted phone or multiview rig for the hotel, laundry, apparel and front-of-house rows. 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 garment work. Force and tactile readings are not among the streams, no thermal or temperature stream is among the ten, and no robot or teleoperation data is captured.

Each attempt is delivered as one episode. Inside it, action segments divide the episode 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 a pressing sequence becomes a run of labeled steps such as lay, press, turn and hang rather than one long clip. 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 the two-handed structure of garment work: where one hand holds the fabric still while the other presses, hangs or sorts. The site describes contact events as the moment the hand meets the item, and does not say whether a hanger, a press or the garment itself counts as the item, so the spec is the place to settle what a contact event means for your task.

Masks are the stated answer for fabric. 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. Whether a given garment needs them is settled in the spec rather than assumed.

The site does not list labels for the state of the cloth. There is no stated label for a wrinkle map, a garment-state or fold-state annotation, a grasp point, a hanging success flag, a contact force or pressure reading, or a task-completion score. Catalog descriptions such as pressing seams or spacing hangers evenly describe the task the contributor performs, not a delivered measurement.

Several limits the site already states apply directly to garment 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, and a hand under a sleeve or inside a folded collar is an everyday case here. The second is fabric, where per-frame instance masks are the stated answer and the household dataset notes that deformable objects dominate its annotation cost. The third is depth, which the site says degrades on wet and reflective surfaces and is handled with per-frame confidence maps. The site does not say how depth behaves on fabric.

The fourth 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.

Steam is named on the site only as a cooking condition, where it is flagged at the episode level because it disrupts depth. The pressing rows do not state a steam flag, so this page does not claim one. What the site does not state is just as useful to know before you scope: no mask accuracy figure for garments, no description of how a pressing surface or a steam press is handled, and no cloth-state 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.

Homes add their own conditions. The site states that household capture happens inside lived-in spaces, with the bedroom and utility room among the environments listed for the household chores dataset, and that bystanders are de-identified or the footage is discarded. Garments can carry printed labels, names or logos. The site does not list printed text on clothing as a blur category, so this page does not claim it is blurred. The listed categories are faces, plates and screens. If text on items in frame matters for your use, raise it in the spec so the handling is agreed before capture.

On safety and qualification, the site states no contributor safety protocol, no heat-safety guidance for pressing or ironing, and no safety certification for any garment row, and this page does not describe one. 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. After the first batch, on-demand top-ups run against the same accept criteria, so you can add more garments, steps or settings 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 wrinkles, cloth or fold state, grasp points or hanging success, no force or tactile stream, no garment, fabric, press or hanger specification, and no garment count. Nothing here claims a training outcome or a model improvement.

The tradeoff to plan for is the embodiment gap between a human hand and a robot end-effector, 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 garments, the steps, 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 garment 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 item 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 pressing station, a retail rail and a deformables bench present different geometry and the catalog lists a different capture setup for them. And name whether the task must be a home task: the household dataset lists folding and sorting but not ironing, so a domestic ironing request is settled in the spec.