One knife, one piece
The row names a knife and a block of wood. The knife stays in the hand for the whole task, and the piece is usually held in the other, so the hands are the data.
Now recruiting capture teams worldwide. Get paid to record first-person video of real work.
Apply to captureWhittling
Why whittling needs its own spec
The row names a knife and a block of wood. The knife stays in the hand for the whole task, and the piece is usually held in the other, so the hands are the data.
Whittling is rated Medium, while the neighbouring wood carving row is rated Hard, so a spec should say which one it wants and at what level of detail.
The row says small figures and objects carved from a block, so the piece changes shape across the episode. The spec should say what the manipulated object is at each step.
The row is available on request, which the site defines as scoped and recruited to spec when a buyer asks for it.
Definition
Egocentric whittling data is first-person video of one person carving a small figure or object from a block of wood with a knife, recorded from that person’s own point of view while they work. A paid, consented contributor wears the capture rig and performs the task written in a skill spec, and each attempt is delivered as an episode with synchronized streams, hand pose and labeled action segments. It is human capture. It is not robot or teleoperation data, and it is not a fixed corpus you download.
The site states the task in its commercial task catalog. In the Woodcraft & Carving Studio environment, in the Creative Workshops & Crafts sector, the row "Whittling" reads: "Carve small figures and objects from a block of wood with a knife." It is rated Medium, captured with a head-mounted phone or a multiview rig, and listed as available on request. This page is built on that row and on nothing else the site says about whittling.
People reach this topic through egocentric whittling data, first-person knife carving video, a small wood figure carving dataset or custom woodcraft data collection.
Two neighbouring pages are easy to confuse with it. The egocentric wood carving page covers the "Wood carving" and "Whittling" rows together and is organised around carving as a whole. The stereo woodcraft and carving page covers all five studio rows with a stereo and container focus. This page takes the single whittling row and treats it in depth.
Whittling task
The first table below reproduces the five catalog rows of the Woodcraft & Carving Studio environment, with the whittling row first. The other four, "Wood carving", "Woodturning", "Marquetry" and "Furniture crafting", are different sequences and are listed for context only. The environment is described with examples of woodturning, wood carving, marquetry, furniture crafting and whittling.
The whittling row names one action, carve, one tool, a knife, and one starting material, a block of wood. It names small figures and objects as the result. The site does not say which figure or object, which knife, what wood, how large the block is, how many cuts the piece takes or what the finished piece looks like. Those are spec decisions, and a Spec Pilot is the place to settle them.
The row is listed as available on request with the head-mounted phone or multiview rig capture setup. It is not listed as ready to deploy or as an active collection, so expect a scoping step before capture starts.
Neighbouring pages
The catalog separates whittling from wood carving. "Wood carving" is rated Hard and names chisels, gouges and knives, while "Whittling" is rated Medium and names a knife only. The egocentric wood carving page covers both rows and is the right starting point if you want the two compared or collected together. This page is for a request that is only about the knife and the small piece.
The stereo woodcraft and carving page covers all five studio rows with the stereo pair and the MCAP container in focus. Choose it when depth and a single log file are the point. The egocentric marquetry page covers cutting veneer pieces to pattern and gluing them into an inlay, and the woodturning pages cover the lathe. Those are different tools and different sequences, and none of them is built around the whittling row alone. The comparison table below sets the pages side by side.
Episodes
Each attempt is delivered as one episode, and whittling needs a decision up front about what one attempt is. It could be one cut, one stage such as roughing out a form or adding detail, or the whole figure from block to finished piece. The site does not set an episode length, the size of the piece or the number of episodes, so the skill spec does. Inside the episode, the synchronized streams run from start to end, and action segments divide it into labeled spans.
The choice matters because of two machine-checked rules. An episode is rejected if the manipulated object is out of frame for more than 15 percent of it, and if more than 35 percent of depth pixels in the manipulation region are invalid. In whittling the piece is small and the tool and the hands are all working on it, so the spec should say what counts as the manipulated object and where the manipulation region is at each step.
The site does not describe how a piece held in the hand, a fast-moving tool or loose material is handled. If your policy depends on any of those, say so at the start, and settle it in a Spec Pilot rather than assuming it.
Labels and delivery
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 a segment can be one step as written in the skill spec, so a whittling session becomes a run of labeled steps 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 how the hands grip the knife and the piece and how they move together, and it is the main signal for the manipulation inside a whittling task. The site describes contact events as the moment the hand meets an item, and does not say which contacts matter in whittling, so the spec settles what a contact event means for your task.
The site lists no label for the outcome of the carving. There is no stated figure type, wood species, cut depth, dimension, surface quality, finish label or task-completion score, and no contact force reading. Camera pose, a head trajectory and a map of the workspace are likewise not among the listed deliverables. The site states 6-DoF IMU data at 200 Hz from head- and wrist-mounted sensors, and nothing more about motion.
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. The official documentation for three of these formats is linked in the sources below. Collections are billed per validated hour, and the license is perpetual, irrevocable and worldwide, buyer-owned and non-exclusive by default, with exclusivity available as a separate term.
Why it matters
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 hand work sits in that base layer. Human capture does not replace robot data, and this page does not claim that it does.
Whittling is a useful shape of task because the work is in the fingers. The site’s tool use page says that driver and bit selection, grip changes and tool hand-off between hands live in finger-level detail that RGB video alone cannot resolve, and that each needs 3D hand pose tracked through reach, grasp and tool manipulation. That page is about driving and fastening tools, not knives, so this page does not claim the same findings for whittling. What carries over is the reasoning: a clip that shows a piece somewhere in frame does not carry what a policy needs. The hands, the knife and the step boundaries have to be in the episode.
A custom collection lets you aim at the step your model is missing, such as roughing out a form versus adding detail, or a specific grip on the knife. This page does not claim that any public corpus lacks whittling data, and it does not claim a training outcome or a model improvement from any dataset.
The tradeoff to plan for is the embodiment gap between a human pair of hands whittling and a robot platform doing the same, which retargeting or pairing with robot data has to close. Counts of contributors, hours and tasks are set in your skill spec.
Capture setup
The catalog lists "Phone (head-mounted) / multiview rig" for the whittling row. It does not list a stereo pair for it, so this page does not describe whittling as stereo capture. Stereo applies only through the site’s general stereo offer: a stereo pair is a capture choice agreed in the skill spec and confirmed on a Spec Pilot.
The site states that a stereo pair at human-eye spacing makes depth recoverable from every frame pair, and that its head-mounted stereo rig carries a 4K head camera and two wrist cameras. It gives no depth accuracy figure for whittling and does not say how depth behaves on a small piece held in the hand. It does say that depth degrades on wet and reflective surfaces, and that a confidence map lets you mask low-confidence regions.
The handheld-gripper setup, which the site describes as a staged working set, is not the setup listed for this row. The site names no studio, school, workshop, city or region, and it lists capture at your own facility as not offered. This page therefore makes no claim that episodes are recorded in a particular studio, and none that they are recorded at yours.
Privacy
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. An undocumented bystander is also one of the machine-checked reject criteria. The anonymization record is delivered with the data.
On safety and qualification, the site states no contributor safety protocol, no injury-prevention guidance and no certification for the whittling row, and this page describes none. If your program needs terms about who performs the work 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.
Speed and delivery
From signed spec to first validated batch, the site states 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. The site does not say whether the Woodcraft & Carving Studio environment counts as deep or thin, and its rows are available on request, so ask which timeline applies when you request the quote. Speed comes from a warm pipeline, not from cutting consent, review or anonymization.
The site also states program milestones: a pilot reaches its first delivery in two to three weeks, and a standing program reaches full rate in four to six weeks. After the first batch, on-demand top-ups run against the same accept criteria, so you can add pieces, steps or grips without rewriting the spec. The site describes its custom collection as custom collection at scale through a network of thousands of capture providers, for a consistent, recurring data flow to labs. It does not state a turnaround for later batches, so plan them in your quote.
Only footage that clears the signed accept criteria is billed as a validated hour, and rejected items are recaptured at our cost. The site does not publish a price list, a minimum order or a batch size for this task, so those are set in your quote.
Not stated
Where this site does not state something, this page says "Not stated" rather than filling the gap. The tables below separate what is delivered from what is not stated. There is no listed label for figure type, wood species, cut depth, surface quality or completion, no knife specification, no task duration, no site count, no region or city, no episode count and no success or quality label. Nothing here reports a robot result, a benchmark or a model improvement.
Counts of contributors, hours and tasks are set in your skill spec. A task outside the row quoted above, such as a carving technique the catalog does not name, is settled in a Spec Pilot rather than assumed.
Getting started
Describe the step, the piece 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, a small, fast first batch that proves the spec and the accept criteria on real footage before volume scales. The scoping checklist below maps the parts of a skill spec onto a whittling 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. Name the step and the piece rather than a broad activity, because that is what lets each step become one action segment. Decide whether you want a stereo pair through the general stereo offer, since the catalog lists none for this row. And name where the episode starts and stops: one cut, one stage or the whole figure, since the site does not set an episode length.
At a glance
Every collection runs through the same seven-stage end-to-end custom collection pipeline. End-to-end custom data collection is a managed service that takes an AI data need from problem to owned dataset in a single accountable pipeline.
Woodcraft rows
| Task (as listed) | Description (as listed) | Environment | Skill level | Capture setup | Status |
|---|---|---|---|---|---|
| Whittling | Carve small figures and objects from a block of wood with a knife. | Woodcraft & Carving Studio | Medium | Phone (head-mounted) / multiview rig | Available on request |
| Wood carving | Shape and detail wood with chisels, gouges and knives into a carved piece. | Woodcraft & Carving Studio | Hard | Phone (head-mounted) / multiview rig | Available on request |
| Woodturning | Mount a blank on the lathe and shape it with turning tools as it spins. | Woodcraft & Carving Studio | Hard | Phone (head-mounted) / multiview rig | Available on request |
| Marquetry | Cut veneer pieces to pattern and glue them into a decorative inlay on a panel. | Woodcraft & Carving Studio | Hard | Phone (head-mounted) / multiview rig | Available on request |
| Furniture crafting | Cut, join and finish timber components into furniture pieces at the bench. | Woodcraft & Carving Studio | Hard | Phone (head-mounted) / multiview rig | Available on request |
Page comparison
| Page | Main topic | Catalog rows it covers | Choose it when |
|---|---|---|---|
| Egocentric whittling video data (this page) | The knife and the small piece | "Whittling" (Medium) | Your request is only about whittling |
| Egocentric wood carving video data | Carving and whittling together | "Wood carving" (Hard) and "Whittling" (Medium) | You want the two compared or collected together |
| Stereo woodcraft and carving data | The stereo pair and the MCAP container for woodcraft | All five studio rows | Depth and a single MCAP file are the point |
| Egocentric marquetry and inlay data | Cutting veneer to pattern and gluing it into an inlay | "Marquetry" (Hard) | The task is inlay, not carving |
| Egocentric woodturning and lathe data | Shaping a blank on the lathe | "Woodturning" (Hard) | The task is lathe work |
Streams and labels
| Stream or label | What it gives a whittling episode | Status |
|---|---|---|
| Head-mounted stereo rig, 4K head camera and two wrist cameras | First-person view of the task plus a closer view of the hands, where a stereo pair is agreed in the spec | Delivered where specified |
| Depth with confidence maps | Scene geometry, with a confidence map to mask low-confidence regions; degrades on wet and reflective surfaces | Delivered |
| 6-DoF IMU at 200 Hz | Head and wrist motion aligned to the video frame | Delivered |
| Hardware sync | Ten streams on one PTP clock domain, aligned under 2 ms | Delivered |
| 3D hand and body pose, with per-joint visibility flags for the hands | Both hands across the step; occluded fingers flagged | Delivered |
| Contact events | The moment the hand meets an item | Delivered |
| Action segments (start_ns, end_ns, verb, noun) | One labeled span per step as written in the spec | Delivered |
| Per-frame instance masks (848×480) | Masks where the spec calls for them | Delivered where specified |
| Consent artefact IDs and anonymization record | Provenance per episode | Delivered |
| Figure type, wood species, knife, cut depth, surface quality or completion label | Not listed | Not stated |
| Camera pose, head trajectory or workspace map | Not among the listed deliverables | Not stated |
| Contact force or tactile readings | Not among the listed streams | Not stated |
| Safety or certification | Not among the listed streams | Not stated |
| Robot or teleoperation data | Human first-person demonstrations only | Not captured |
| Counts of contributors, hours, tasks and episodes | Set per program in the skill spec | Not stated |
Delivery formats
| Format | What the official documentation states | Official source |
|---|---|---|
| RLDS | An ecosystem of tools to store, retrieve and manipulate episodic data; a dataset is a set of episodes, each holding steps and metadata | RLDS repository (GitHub) |
| LeRobot | LeRobotDataset v3.0 stores many episodes per Parquet and MP4 file, with episode boundaries, lengths and tasks resolved through metadata | LeRobot dataset v3 docs (Hugging Face) |
| WebDataset | Tar files in which files that share a basename make up one training sample, split into numbered shards | WebDataset repository (GitHub) |
| HDF5, zarr, Rerun .rrd, raw streams with metadata | Listed as delivery options on this site; not documented on this page | Not stated here |
Scoping checklist
| Skill-spec part | What to name for a whittling request | Drawn from what the site states |
|---|---|---|
| Named skill | The verb and the noun for each step, and the tool: carve a form with a knife, add detail with a knife. Not "whittling" | Catalog wording: "Carve small figures and objects from a block of wood with a knife." |
| Geometry that must be visible | What must be in frame and for how long: both hands, the knife and the piece at each step | Set in the skill spec |
| Condition budget in hours | Hours per condition cell, including pieces, grips and failure cases | Set per program in the skill spec |
| Reject rule | The conditions that fail an item, written first | 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 |
| Acceptance test | One test that decides whether an attempt counts as complete | Your acceptance test is agreed in the spec |
| Setting and capture tier | Woodcraft & Carving Studio, and phone or multiview rig, with a stereo pair only if agreed | The catalog lists a capture setup per row; stereo applies through the general stereo offer |
| Episode boundary | One cut, one stage or the whole figure | Set in the skill spec; the site does not state an episode length |
Example footage
One real clip already published on this site, shown as a format and rig example: a head-mounted stereo capture with hands in frame. It does not show whittling, wood carving or any woodcraft task.
No whittling or woodcraft clip is published on the site, so this is labelled as a format and rig example. Request a sample to see delivered files. See stereo sample page, stereo + wrist-cam example videos, request an egocentric data sample.
FAQ
Sources
Tell us the step, the piece and the variation or failure cases your pipeline needs covered. We will turn it into a written skill spec and run a Spec Pilot before scaling to a full collection.