Now recruiting capture teams worldwide. Get paid to record first-person video of real work.

Apply to capture

Task family

Stereo Pottery Wheel Throwing and Hand-Building Data

Centring clay on the wheel, opening and pulling walls and trimming the finished form, or rolling slabs and coils and joining them by hand into a vessel or form: these are the forming steps of a ceramics studio, done with wet, deformable material that hides much of what the fingers are doing. This page covers wheel throwing robot training data and a stereo pottery wheel video and egocentric hand-building dataset for the two catalog rows that name this work: what a calibrated stereo pair adds, what MCAP is according to its own documentation, and what the site does not state. We run custom ceramics data collection fast and on demand against a spec you write, through custom collection at scale through a network of thousands of capture providers, for a consistent, recurring data flow to labs.
Catalog rows
Two of five, in Pottery & Ceramics Studio
Skill and status
Wheel throwing Hard, Hand-building Medium; both Available on request
Container
MCAP (.mcap), open source
License
Perpetual, buyer-owned

Why stereo and MCAP here

What do a stereo pair and an MCAP file add to wheel throwing and hand-building video?

01

Depth for a form that changes shape

A thrown wall rises, thins and flares while the wheel turns, and a hand-built form grows join by join. Two calibrated cameras give depth alongside the colour image, which a single view cannot give for a shape that is changing under the fingers.

02

Fingers in the clay, flagged when hidden

The fingers work directly on the form, and clay often hides the part being worked. 3D hand pose with a visibility flag per joint says which joints were seen and which were not, rather than guessing silently.

03

Two different task shapes on one scoped order

Wheel throwing is one continuous sequence on a wheel. Hand-building is a rolling, cutting and joining sequence with no wheel named. Both sit in one catalog environment, so one written spec can carry both, with each sequence defined and labelled separately.

04

Several streams in one container

MCAP stores multiple channels of timestamped, pre-serialized data, so a stereo pair, hand pose and IMU can share one file on one log-time axis. Which streams go in the file is set in the skill spec.

05

Rights you can ship on

A perpetual, buyer-owned commercial license backed by a documented consent chain, so a policy trained on the data can go into a product.

Definition

What is stereo pottery wheel throwing and hand-building data?

It is real-world, first-person video of a person doing the work the two catalog rows describe: centring clay on the wheel, opening and pulling walls and trimming the finished form, and rolling slabs and coils and joining them by hand into a vessel or form. Stereo describes the capture, with two calibrated cameras, and MCAP describes the container. They are separate choices, and this page covers both for this work.

People reach this topic through several phrasings: wheel throwing robot training data, a stereo pottery wheel video, an egocentric hand-building dataset, or custom ceramics data collection at scale. They describe one request from different angles. The data is human capture by paid, consented contributors, not robot or teleoperation data, so using it for a specific robot still needs your own retargeting step.

No named robot or model has been trained on Firsthand data for this work, and Firsthand is not affiliated with Foxglove or the MCAP maintainers.

Catalog rows

Which catalog rows cover wheel throwing and hand-building?

The site lists two rows in its Pottery & Ceramics Studio environment, in the Creative Workshops & Crafts sector. "Wheel throwing" reads "Centre clay on the wheel, open and pull walls and trim the finished form." and is rated Hard. "Hand-building" reads "Roll slabs and coils and join them by hand into a vessel or form." and is rated Medium.

Both rows list their capture mode as Phone (head-mounted) / multiview rig, and both are listed as Available on request. The environment holds three further rows, ceramic sculpting, glazing and kiln loading and firing, which this page does not claim.

The site names no clay body, wheel model, tool or brand for these rows, so those details are agreed in the written spec.

Status

What does Available on request mean for these rows?

The site defines Available on request plainly: the work is scoped and recruited to spec when a buyer asks. It is not the same as Ready to deploy, where contributors and site access are already set up, and it is not Active collection, where recording is already running and a spec can join an existing stream.

That means recruitment comes first for a wheel throwing or hand-building order, and the timeline is quoted against your spec. The site does not state a volume, a number of studios or a number of hours for either row, and none are claimed here. It also states no delivered stereo footage for these rows.

The site lists no off-the-shelf pottery dataset, so this work is custom collection.

Stereo

Is this data captured in stereo, and what does a stereo pair add?

Not on the rows as listed. The catalog names a head-mounted phone or multiview rig as the capture mode for both, so a stereo pair is a capture choice agreed in the skill spec through the site general stereo offer and confirmed on a Spec Pilot. This page claims no delivered stereo footage of wheel throwing or hand-building.

Where a pair is agreed, two calibrated cameras record the same scene from two viewpoints, and depth is derived alongside the image. For a thrown wall or a coil being joined, that gives a record of how the form sits in space that one camera cannot supply. The site states no accuracy figure for this work, and none is claimed here.

For the stereo head rig listed under dataset 06, the site states dual fisheye cameras at 1080p per eye on a 62 mm baseline, per-unit calibration, a 9-axis IMU and hand-tracking output, shipped as MCAP. Whether a pottery order uses that rig is not stated, so the rig is set in the spec and confirmed on a Spec Pilot batch.

MCAP

What is MCAP, and why would a wheel throwing order use it?

According to its official site, MCAP is an open source container file format for multimodal log data. It supports multiple channels of timestamped, pre-serialized data, so several streams can sit in one file on one time axis. The official repository, foxglove/mcap, carries the libraries and command line tool and is MIT licensed. Both are linked in the sources below.

For this work, a stereo pair, hand pose and IMU readings could share one file, which keeps the streams of a throwing session together. MCAP is a container, not a training format. If your training loop expects RLDS, LeRobot or WebDataset, name that format in the spec alongside MCAP.

Topic names, message encoding and rates for a pottery order are not stated on the site and are agreed in the skill spec. This page invents none.

Wheel throwing

What does the wheel throwing row involve?

The row describes one continuous sequence: centre clay on the wheel, open and pull walls, and trim the finished form. An episode can cover the whole sequence or isolate one part of it, for example only the centring, only the pulling of walls or only the trim.

The site rates the row Hard, one of two Hard rows in the environment. It gives no clay type, wheel model, wheel speed, piece size, number of pieces per episode or task time for this row, and none are claimed here. Which form is thrown, where an episode begins and ends and whether the trim counts as part of the same episode are agreed in the spec rather than assumed.

Hand-building

What does the hand-building row involve?

The row describes forming without a wheel: roll slabs and coils and join them by hand into a vessel or form. An egocentric hand-building dataset therefore centres on rolling, cutting and joining pieces of clay with the hands, as the potter sees them.

The site rates the row Medium. It states no piece size, number of slabs or coils, build time or tool list for this row, and it adds no instruction beyond the wording of the row. Whether an episode is a single join, one coil laid or a whole build is agreed in the spec.

Two rows compared

How do wheel throwing and hand-building differ as data?

They differ in what the catalog says about them. Wheel throwing is rated Hard and names a wheel; hand-building is rated Medium and names slabs and coils. Both share the same capture mode, the same status and the same environment, so one written spec can carry both while keeping each sequence separately defined and labelled.

The step vocabulary is different too. Throwing steps in the row are centring, opening, pulling walls and trimming. Hand-building steps in the row are rolling slabs and coils and joining them by hand. Action segments carry a verb and a noun, so your spec decides how finely each is cut. The comparison table below sets the two rows side by side as the catalog states them.

Hard cases

What goes wrong when you record wheel throwing and hand-building?

Hands are the first risk. The fingers are inside or around the clay while the form is opened, pulled or joined, and they hide the part being worked. The visibility flag on each joint is the answer the site gives, contact events mark when a hand meets an item, and masks are recorded per frame. Whether hand pose holds while the hands are in wet clay on a turning wheel is not stated, so a throwing order should check it on a Spec Pilot.

Surfaces are the second. The site states that depth degrades on wet and reflective surfaces, with confidence maps masking low-confidence regions, and that an episode is rejected when more than 35 percent of depth pixels in the manipulation region are invalid. Whether wet clay counts as a wet surface for this purpose is not stated, and it belongs in the pilot check.

Shape is the third. The site lists no force or tactile stream and no label for how a form comes out, so a spec that needs a shape outcome or a pressure reading has to define it.

Streams and labels

What does each episode contain?

Each attempt is delivered as one episode, following the episode schema used across Firsthand custom collections: 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. Each episode carries one task instruction. Hand pose is stated as 21 joints per hand with a per-joint visibility flag, and sync is hardware-triggered onto one clock with a 2.0 ms cross-stream ceiling.

Action segments divide an episode into start- and end-timestamped spans, each carrying start_ns, end_ns, verb and noun. For this work, a segment might be one centring, one wall pulled, one slab rolled or one coil joined, as written in the skill spec. The site lists no label for the outcome of a step, such as whether a wall came out even.

Neighbouring pages

How is this different from the other pottery and ceramics pages?

The egocentric pottery wheel throwing and hand-building page covers the same two rows as first-person capture and does not make stereo or MCAP its subject. This page is the stereo and MCAP version. The stereo pottery and ceramics page covers all five pottery rows in one page, and the stereo pottery glazing and kiln loading page covers the other two rows that the site pairs together.

The egocentric pottery and ceramics page covers all five rows without stereo as its subject. None of the neighbours makes wheel throwing and hand-building together its main stereo topic. The comparison table below sets each against this page.

Privacy

How is contributor privacy handled?

Every contributor signs a written release before capture, and each delivered episode links to that record through a documented consent chain. Contributors are paid and consented, and footage that cannot be de-identified is discarded rather than shipped.

At ingest, automated detection finds faces, plates and screens and blurs them irreversibly, and a reviewer confirms the result before anything ships. The anonymization record is delivered with the data, and an undocumented bystander is a stated reject criterion. Whether finished pieces or original designs are blurred or protected beyond the license is not stated. None of this is legal advice, and the executed agreement controls.

Delivery

How is the data labeled and delivered?

Delivery formats are RLDS, LeRobot and WebDataset, plus HDF5, zarr and Rerun .rrd, and a custom order can be packaged as MCAP. The first three are training-loader formats, HDF5 and zarr carry arrays for custom pipelines, and Rerun .rrd gives a multi-stream synchronized playback file for review. The format table below sets each against its official documentation where it was read.

Every batch is validated against signed accept criteria before it ships, with a reject log, and delivery is rolling, per validated batch. Object labels use the terms your spec sets, so labels follow the vocabulary your training stack expects rather than a generic taxonomy.

Fast · on demand · at scale

How fast can a custom wheel throwing and hand-building dataset be collected?

Where coverage is already deep, the first validated batch lands in a median of 48 hours from signed spec. Thin environments typically take three to four weeks, because vetting operators there is the bottleneck, and we quote that timeline up front. Both rows are Available on request, so recruitment comes first and the range that applies is confirmed when the spec is quoted.

On-demand top-ups run against the same spec and accept criteria after each training run, so new 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.

This is 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 counts, hours or throughput for this task, and none are claimed here.

The spec

How do you start a pilot?

You describe the throwing and hand-building sequences your policy needs. 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.

A Spec Pilot, a small and fast first batch, can prove the spec, the stereo choice and the accept criteria before you scale volume. The collection runs through the same seven-stage end-to-end custom collection pipeline as every Firsthand program: contributor recruiting, capture, per-batch QA, anonymization and licensed delivery. Billing is per validated hour, a reject log ships with every batch, and rejected items are recaptured at our cost.

At a glance

What you get.

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.

What the site states

What does the site state about stereo wheel throwing and hand-building?

Where the work appears on this site, what each place states and how this page uses it.
Where it appearsWhat the site statesUsed on this page as
Task catalog, Pottery & Ceramics StudioFive rows; two are Wheel throwing (Hard) and Hand-building (Medium), both Available on requestThe stated tasks
Task catalog capture modePhone (head-mounted) / multiview rig on both rows; no stereo pair is stated for themHow the rows are listed
General stereo offerA stereo pair is agreed in the skill spec and confirmed on a Spec PilotHow stereo applies, if you want it
Stereo head rig (06)Dual fisheye, 1080p per eye, 62 mm baseline, per-unit calibration, 9-axis IMU, hand-tracking output; shipped as MCAPThe stated stereo rig; whether a pottery order uses it is not stated
Episode schemaTen streams synced under 2 ms; one task instruction per episodeWhat each episode carries
Task catalogNo clay type, wheel model, piece size, task time or success rate is statedNot stated; scoped in the written spec

Catalog rows

Which two catalog rows does this page cover?

The two rows exactly as listed in the Pottery & Ceramics Studio environment, with the catalog description, skill rating, capture mode and status.
Task familyCatalog task and descriptionSkillCapture modeStatus
ThrowingWheel throwing: centre clay on the wheel, open and pull walls and trim the finished formHardPhone (head-mounted) / multiview rigAvailable on request
Hand-buildingHand-building: roll slabs and coils and join them by hand into a vessel or formMediumPhone (head-mounted) / multiview rigAvailable on request

Two rows compared

How do the Wheel throwing and Hand-building rows compare?

The two rows side by side, using only what the catalog states and where the site is silent.
AspectWheel throwingHand-building
Catalog wordingCentre clay on the wheel, open and pull walls and trim the finished formRoll slabs and coils and join them by hand into a vessel or form
Skill ratingHardMedium
StatusAvailable on requestAvailable on request
Capture modePhone (head-mounted) / multiview rigPhone (head-mounted) / multiview rig
Steps named in the rowCentre, open, pull walls, trimRoll slabs, roll coils, join by hand
Equipment named in the rowThe wheelNone beyond the hands
Clay type, wheel speed, piece size, task timeNot statedNot stated
Where step boundaries are setThe written specThe written spec

Capture setup

How is this work recorded, and where does stereo apply?

The capture mode named in the catalog for the two rows, and the stereo choices the site states.
Capture setupWhat the site says it recordsWhere it applies
Phone (head-mounted) / multiview rigA head-mounted phone records the first-person view of real work; a multiview rig adds synchronized third-person angles when the spec asks for themBoth catalog rows
Stereo pairTwo calibrated cameras with depth alongside; agreed in the skill spec and confirmed on a Spec PilotThrough the general stereo offer, not on the rows as listed
Stereo head rig (06)Dual fisheye, 1080p per eye, 62 mm baseline, per-unit calibration, 9-axis IMU, hand-tracking outputWhether a pottery order uses it: not stated
Wrist-mounted cameraAvailable alongside the head camera for close-range detailStated as an option; per-task use set in the spec
Hand pose21 joints per hand with a per-joint visibility flagAll episodes
SyncHardware-triggered onto one clock; 2.0 ms cross-stream ceilingAll episodes
Customer facilityCapture at a customer facility is not offeredStudio and contributors agreed in the spec

Streams and labels

What does each episode carry, and what is not stated?

Streams and labels the site states for Firsthand custom collections, and fields the site does not state for these rows.
ItemStatus on this siteWhy it matters here
Head and wrist RGB, depthStated; ten streams synced under 2 msKeeps the clay and both hands on one timeline
3D hand pose with per-joint visibilityStated; 21 joints per handCarries how the fingers sit on the form while it is opened, pulled, rolled or joined
Contact and grasp eventsStatedMarks when the fingers make and break contact with the clay
Per-frame instance masksStated; stored at 848×480Separates the piece from the wheel, the bench and the hands
Action segmentsStated; frame-accurate, with start_ns, end_ns, verb and nounMarks centring, opening, pulling, trimming, rolling and joining as separate steps
Task instructionStated; one per episodeNames the sequence the episode shows
MCAP topics, schemas, ratesNot stated on the siteAgreed in the skill spec
Force or tactile dataNot listedDefined in the spec if the policy needs them
Clay types, wheel models, piece sizesNot statedDefined in the spec
Shape-outcome label, success rateNot statedNo such claim is made on this page

Formats

Which formats carry this data, and where does MCAP fit?

Delivery formats stated on the site, set against what each format official documentation states where the documentation was read.
FormatWhat the official documentation statesUsed for here
MCAPAn open source container for multimodal log data with multiple channels of timestamped pre-serialized data (mcap.dev)Multi-stream timestamped delivery of a stereo pair, hand pose and IMU, with layout set in the spec
RLDSA dataset is retrieved as episodes, and each episode holds a dataset of steps (RLDS repository README)Episode-based training loaders
LeRobotLeRobotDataset v3.0 is a standardized format for robot learning data, including multi-camera video, with many episodes stored per Parquet or MP4 file (LeRobot documentation)LeRobot dataset format and tooling
WebDatasetTar files in which files that belong together share a basename, split into numbered shards (WebDataset README)Sharded streaming for large training runs
HDF5 and zarrStated by Firsthand onlyArray storage for custom pipelines
Rerun .rrdStated by Firsthand onlyMulti-stream synchronized playback for review

Related pages

How is this page different from the neighbouring pottery and stereo pages?

Related collection pages on this site and how their main topic differs from this one.
PageMain topicHow this page differs
Stereo pottery wheel throwing and hand-building data (this page)The two Pottery & Ceramics Studio rows for wheel throwing and hand-building, with stereo and MCAP framingCentres on both forming tasks together as a stereo order
Egocentric pottery wheel throwing and hand-building dataThe same two rows in first-person capture, without stereo or MCAP as its subjectThis page adds the stereo pair and the MCAP container
Stereo pottery and ceramics dataAll five Pottery & Ceramics Studio rows with stereo and MCAP framingGlazing, kiln loading and ceramic sculpting are in scope there, not here
Stereo pottery glazing and kiln loading dataThe glazing and kiln loading and firing rows with stereo and MCAP framingA different pair of rows: surface and firing, not forming
Egocentric pottery and ceramics video dataAll five Pottery & Ceramics Studio rows in first-person captureCovers the whole environment, with stereo not its subject
Egocentric stereo data in MCAP formatThe stereo and MCAP offer across tasksA general format page, not a pottery page

Skill spec

What the skill spec fixes for this order

Spec element and where a buyer can see it in the delivery. Set per program means the site states no fixed value.
Spec elementWhat it decidesWhere it shows up
Tasks and orderWhether an episode is throwing, hand-building or both, and where each starts and endsEpisode boundaries and action segments; set per program
Forms and toolsWhich forms are thrown or built and which tools are in scopeObject labels; set per program
Stereo pairWhether a calibrated stereo pair is used, and on which viewsThe channels you can read; confirm on the Spec Pilot
MCAP layoutWhich streams go in the file, and their topic names and ratesThe .mcap file; set per program
Wet clay and depthWhether wet clay is in scope and how depth is checkedSpec Pilot result; set per program
Step labelsWhether centre, open, pull, trim, roll and join are separate labelled stepsAction segments; set per program
Delivery formatHow the data is packagedRLDS, LeRobot, WebDataset, HDF5, zarr, Rerun .rrd or MCAP
Accept criteriaWhat passes QA, including what counts as a rejectReject log per batch
PrivacyHow people, plates and screens in frame are protectedFaces, plates and screens blurred at ingest
LicenseWho owns the datasetPerpetual, buyer-owned commercial license; exclusivity available

Example footage

What does egocentric stereo capture look like?

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 pottery, clay, a wheel or hand-building.

egocentric_stereo_sample.mp4 · head-mounted stereo · hands in frame · 1920 × 1080 web preview · rig example, not a pottery clip

No wheel throwing or hand-building 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

Wheel throwing robot training data, answered.

What is stereo pottery wheel throwing and hand-building data?
It is first-person video of a person throwing clay on the wheel and building forms from slabs and coils, recorded with two calibrated cameras where a stereo pair is agreed in the spec, with depth and hand pose alongside. Firsthand collects it to a written spec.
Which catalog rows cover it, and what is their status?
Two rows in Pottery & Ceramics Studio: Wheel throwing (Hard) and Hand-building (Medium). Both are Available on request, with a head-mounted phone or multiview rig as the listed capture mode.
Is the data already captured in stereo?
Not on the rows as listed. Stereo applies through the site general stereo offer, agreed in the skill spec and confirmed on a Spec Pilot, and this page claims no delivered stereo footage of these tasks.
What is MCAP and can this data be delivered as MCAP?
MCAP is an open source container for multimodal log data with multiple channels of timestamped, pre-serialized data, according to mcap.dev. A custom order can be packaged as MCAP, with topic names, encoding and rates agreed in the skill spec.
How is this different from the egocentric wheel throwing page?
That page covers the same two rows as first-person capture. This page adds the stereo pair and the MCAP container as its subject, and it is a framing of the same catalog rows, not a different set of tasks.
Does the data include force readings or shape-outcome labels?
Not stated. The site lists no force or tactile stream and no label for how a form comes out. It states hand pose, depth, contact events and action segments.
How fast can a custom dataset be collected?
Where coverage is deep, the first validated batch lands in a median of 48 hours from signed spec. Thin environments typically take three to four weeks, and both rows are Available on request, so recruitment comes first and a Spec Pilot can prove the spec.
What license does the data ship under?
A perpetual, buyer-owned commercial license, with exclusivity available, documented alongside the consent chain.

Sources

Where are MCAP and the delivery formats documented?

Request a custom stereo wheel throwing and hand-building dataset.

Tell us the forms and tools, whether an episode is throwing, hand-building or both, whether wet clay is in scope, whether you want a stereo pair, the streams you want in the MCAP file and the delivery format. We will quote reach, timeline and price against a written spec, with a first validated batch in a median of 48 hours where coverage is deep.