Hands, tools and the piece in one close view
A head-mounted view keeps both hands, the pipe or the rotary tool, and the piece in frame while the glass is masked, etched, gathered or shaped.
Now recruiting capture teams worldwide. Get paid to record first-person video of real work.
Apply to captureTask family
Requirements
A head-mounted view keeps both hands, the pipe or the rotary tool, and the piece in frame while the glass is masked, etched, gathered or shaped.
3D hand pose with a per-joint visibility flag records where the fingers are, and which joints were hidden, while a small piece is held close.
Etching is a masked, stepwise surface job. Glassblowing is a continuous sequence at the pipe. Both sit in one catalog environment, so one written spec can carry both, with each sequence defined and labelled separately.
Frame-accurate action segments mark where one step ends and the next begins, aligned to video and depth.
A perpetual, buyer-owned commercial license backed by a documented consent chain, so a policy trained on the data can go into a product.
Why it matters
Egocentric glass etching and glassblowing data is real-world, first-person video of a person doing the work the catalog rows describe: masking the glass, applying etching cream or using a rotary tool and rinsing to reveal a design, and gathering molten glass on the pipe, blowing and shaping it with tools and transferring it to anneal. It is recorded from the worker viewpoint, with depth, 3D hand pose, contact events and frame-accurate action segments aligned on one hardware clock.
The work is a demanding test for a manipulation policy. Pieces are small, the steps are sequential and the hands do most of the work with tools that hide the fingers. A mask has to cover the right area before anything is applied, and a gathered piece has to be turned and shaped in order. Those are the situations where a close first-person view and hand pose help most.
The site already has an egocentric glass and mosaic page covering all four studio rows, and a stereo page for these same two rows. This page does something narrower: it treats glass etching and glassblowing as the subject of an egocentric order, so a team that needs egocentric glassblowing data, first-person glass etching video or a glass studio hand-skill dataset can scope exactly that work.
Catalog rows
The site lists two rows in its Glass & Mosaic Studio environment, in the Creative Workshops & Crafts sector. "Glass etching" reads "Mask the glass, apply etching cream or use a rotary tool and rinse to reveal the design." and is rated Medium. "Glassblowing" reads "Gather molten glass on the pipe, blow and shape it with tools and transfer it to anneal." and is rated Hard.
Both rows list their capture mode as Phone (head-mounted) / multiview rig, and both are listed as Available on request. The environment lists its examples as "Glassblowing studio, stained glass, mosaic tiling, glass etching", and it holds two further rows, mosaic tiling and stained glass assembly, which this page does not claim.
The site names no tool, equipment model, glass type, thickness, temperature or etchant brand for these rows, so those details are agreed in the written spec.
Status
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 glass etching or glassblowing 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.
The site lists no off-the-shelf glass dataset, so this work is custom collection.
Glass etching
The row describes a short chain: mask the glass, apply etching cream or use a rotary tool, and rinse to reveal the design. An episode can cover the whole chain or isolate one part of it, for example only the masking, or only the rinse.
The row offers two routes to the same result, etching cream or a rotary tool, and a spec can ask for one or both. The site gives no mask material, no cream or tool model, no design, no number of pieces per episode and no task time for this row. It also gives no chemical details and no handling instruction beyond the wording of the row, and this page adds none.
The site rates the row Medium. How a design is chosen, where the mask goes and where an episode begins and ends are agreed in the spec rather than assumed.
Glassblowing
The row describes a longer sequence: gather molten glass on the pipe, blow and shape it with tools, and transfer it to anneal. The catalog rates it Hard, and it is the more demanding of the two rows.
The site states no temperatures, anneal times, glass types, tool models or piece sizes for this row. It also lists no thermal stream, so a glassblowing episode carries no temperature reading unless a spec sets one up. This page gives no instruction for working with hot glass beyond the wording of the row.
How the sequence is cut into episodes, what counts as a finished piece and how the transfer to anneal is framed are agreed in the written spec.
One order
They share an environment, a capture mode and a status, and they share a setting: a glass studio where one person works one piece with hand tools. A buyer building a studio craft policy often wants both a surface skill and a shaping skill, and a single spec keeps the vocabulary, the accept criteria and the delivery format the same across them.
Keeping the two sequences separate matters just as much. Etching ends with a design revealed on a piece. Glassblowing ends with a shaped piece transferred to anneal. A spec can ask for them as two episode types, with action segments marking where each step starts and ends.
The site does not state that any episode combines the two, so how often each appears is set in the spec.
The first-person view
The site describes a head-mounted camera as keeping the hands and the work in frame from the viewpoint a manipulation policy acts from. For etching, that means the mask, the cream or rotary tool and the hands sit in the same view. For glassblowing it means the pipe, the gathered piece and the shaping tool do.
Pieces are small and are often hidden by the fingers that hold them. 3D hand pose is stated as 21 joints per hand with a per-joint visibility flag, which says which joints were seen and which were not rather than guessing silently. Contact and grasp events mark when a hand meets an item.
The site states that a head-mounted view is paired, when the spec asks for it, with synchronized third-person angles from a multiview rig, and with wrist cameras for closer detail on the hands. It states no working distance, field of view or resolution specific to these two rows.
Stereo
Not on the rows as listed. The catalog names a head-mounted phone or multiview rig as the capture mode for both, and this page is the egocentric page for these rows, not the stereo one. Stereo applies only through the site general stereo offer: a stereo pair is agreed in the skill spec and confirmed on a Spec Pilot. This page claims no delivered stereo footage of glass etching or glassblowing.
If you want a calibrated stereo pair, depth from a stereo rig and a container such as MCAP for this work, the stereo glass etching and glassblowing page describes that framing. If a head-mounted phone or a multiview rig is enough, this page describes that capture directly.
The site does not state a depth accuracy figure for these rows either way, and a policy that needs one should ask for it in the spec.
Hard cases
Hands are the first risk. Pieces are small and are often hidden by the fingers that hold them, and a pipe or tool can 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.
Surfaces are the second. The site states that wet and reflective surfaces degrade active-stereo depth and that depth confidence maps mask low-confidence regions. How a transparent or reflective piece behaves in depth is not stated, and an etching order in particular should expect that and check it on a Spec Pilot.
Heat is the third. The site notes that no thermal stream is listed, which matters for a row that works with molten glass, and force and tactile readings are not among the listed streams either. A spec that needs any of them has to set them up.
Capture setup
The catalog lists one capture mode for both rows: a head-mounted phone with a multiview rig. That is a first-person view of real work, with a multiview rig adding synchronized third-person angles when the spec asks for them, and wrist cameras giving a closer view of the hands.
The site also states that Firsthand does not capture at a customer facility, and it does not say which glass studio a given program uses. The studio and the contributors are agreed in the skill spec before capture begins.
The site describes a hard-environments stack with a 4K/60 head camera and a 120° lens, and does not say whether a glass order uses it. The capture table below sets what is stated beside what is not.
Streams and labels
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, and per-frame instance masks are stored at the depth resolution of 848×480.
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 etch, as written in the skill spec. The site lists no label for the outcome of a step, such as whether a design came out clean.
Object labels use the terms your spec sets, so labels follow the vocabulary your training stack expects rather than a generic taxonomy.
Neighbouring pages
The egocentric glass and mosaic page covers all four studio rows as first-person craft video, including mosaic tiling and stained glass assembly. Glass etching and glassblowing are two of its four rows, not its subject, and this page takes them as the subject.
The stereo glass etching and glassblowing page covers the same two rows with a stereo and MCAP framing. This page is the egocentric version, built on the catalog capture mode of a head-mounted phone or multiview rig. The stereo glass cutting and forming page covers five glass rows across two environments, and the egocentric glass manufacturing page covers the one heavy industry row.
None of the egocentric pages makes etching and glassblowing its main topic. The comparison table below sets each against this page.
Privacy
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.
Whether finished pieces or original designs are blurred or protected beyond the license is not stated. Safety certifications, protective equipment and injury handling for glass work are not stated either, and this page makes no such claim. None of this is legal advice, and the executed agreement controls.
Delivery
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.
MCAP is a container for timestamped log data, not a training format. If your training loop expects RLDS, LeRobot or WebDataset, name that format in the spec. The site does not state which streams of a glass order map to MCAP channels, so topic names, encoding and rates are agreed in the spec.
Every batch is validated against signed accept criteria before it ships, with a reject log, and delivery is rolling, per validated batch.
Fast · on demand · at scale
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
You describe the etching and glassblowing 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 views 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
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
| Where it appears | What the site states | Used on this page as |
|---|---|---|
| Task catalog, Glass & Mosaic Studio | Two rows: glass etching (Medium) and glassblowing (Hard), both Available on request | The stated tasks |
| Environment examples | Glassblowing studio, stained glass, mosaic tiling, glass etching | Context for the setting |
| Task catalog, same environment | Two further rows: mosaic tiling and stained glass assembly | Neighbouring work, not claimed here |
| Task catalog, capture mode | Phone (head-mounted) / multiview rig for both rows | The egocentric capture setup |
| General stereo offer | A stereo pair is agreed in the skill spec and confirmed on a Spec Pilot | How stereo would apply; not stated on the rows |
| Task catalog | No temperature, anneal time, glass type, tool model or success rate is stated | Not stated; scoped in the written spec |
Catalog rows
| Task family | Catalog task and description | Skill | Capture mode | Status |
|---|---|---|---|---|
| Etching | Glass etching: mask the glass, apply etching cream or use a rotary tool and rinse to reveal the design | Medium | Phone (head-mounted) / multiview rig | Available on request |
| Hot glass | Glassblowing: gather molten glass on the pipe, blow and shape it with tools and transfer it to anneal | Hard | Phone (head-mounted) / multiview rig | Available on request |
Capture setup
| Capture mode | What the site says it records | Where it applies |
|---|---|---|
| Phone (head-mounted) / multiview rig | A head-mounted phone records the first-person view of real work; a multiview rig adds synchronized third-person angles when the spec asks for them | Both catalog rows |
| Head-mounted camera | Keeps the hands, the piece and the tool in frame from the viewpoint a manipulation policy acts from | All episodes |
| Wrist-mounted camera | Available alongside the head camera for close-range detail | Stated as an option; per-task use set in the spec |
| Stereo pair | Applies only through the site general stereo offer: agreed in the skill spec and confirmed on a Spec Pilot | Not stated on the rows |
| Hard-environments stack | 4K/60 head camera, 120° lens; two 1080p/60 wrist cameras; depth 848×480 at 30 fps; 200 Hz 6-DoF IMU | Whether a glass order uses it: not stated |
| Customer facility | Capture at a customer facility is not offered | Studio and contributors agreed in the spec |
Streams and labels
| Item | Status on this site | Why it matters here |
|---|---|---|
| Head and wrist RGB, depth | Stated; ten streams synced under 2 ms | Keeps the piece, the tool and the hands on one timeline |
| 3D hand pose with per-joint visibility | Stated; 21 joints per hand | Carries the grip on a pipe, a rotary tool or a shaping tool |
| Contact and grasp events | Stated | Marks when the fingers make and break contact with an item |
| Per-frame instance masks | Stated; stored at 848×480 | Separates the piece from the pipe, the tool and the bench |
| Action segments | Stated; frame-accurate, with start_ns, end_ns, verb and noun | Marks masking, etching, rinsing, gathering and shaping as separate steps |
| Task instruction | Stated; one per episode | Names the sequence the episode shows |
| Thermal stream, force or tactile data | Not listed | Defined in the spec if the policy needs them |
| Temperatures, anneal times, glass types, tool models | Not stated | Defined in the spec |
| Task-completion label, success rate | Not stated | No such claim is made on this page |
Formats
| Format | What the official documentation states | Used for here |
|---|---|---|
| RLDS | A dataset is retrieved as episodes, and each episode holds a dataset of steps (RLDS repository README) | Episode-based training loaders |
| LeRobot | LeRobotDataset 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 |
| WebDataset | Tar files in which files that belong together share a basename, split into numbered shards (WebDataset README) | Sharded streaming for large training runs |
| MCAP | An open source container for multimodal log data with multiple channels of timestamped pre-serialized data (mcap.dev) | Multi-stream timestamped delivery on a custom order, with layout set in the spec |
| HDF5 and zarr | Stated by Firsthand only | Array storage for custom pipelines |
| Rerun .rrd | Stated by Firsthand only | Multi-stream synchronized playback for review |
Related pages
| Page | Main topic | How this page differs |
|---|---|---|
| Egocentric glass etching and glassblowing data (this page) | The two Glass & Mosaic Studio rows for glass etching and glassblowing, in first-person capture | Centres on both tasks together, egocentric |
| Egocentric glass and mosaic craft video data | All four Glass & Mosaic Studio rows in first-person capture | Mosaic tiling and stained glass are in scope there; here only etching and glassblowing |
| Stereo glass etching and glassblowing data | The same two rows, with stereo and MCAP framing | Stereo and container framing; this page is the egocentric version |
| Stereo glass cutting and forming data | Five glass rows across Heavy Industry & Fabrication and the studio, with stereo and MCAP framing | Glass handling is its subject; this page covers etching and glassblowing only |
| Egocentric glass manufacturing data | The one Glass manufacturing row in first-person capture | Industrial glass work, not studio etching or glassblowing |
| Stereo window and glass cleaning data | Cleaning interior and exterior glass with applicators and squeegees | A cleaning row, not etching or shaping |
Skill spec
| Spec element | What it decides | Where it shows up |
|---|---|---|
| Tasks and order | Whether an episode is etching, glassblowing or both, and where each starts and ends | Episode boundaries and action segments; set per program |
| Pieces and tools | Which pieces are etched or blown and which tools are in scope | Object labels; set per program |
| Views and streams | Head and wrist views, a second angle, hand pose and any others | The channels you can read; confirm on the Spec Pilot |
| Reflective and clear pieces | Whether transparent or reflective pieces are in scope and how depth is checked | Spec Pilot result; set per program |
| Step labels | Whether mask, etch, rinse, gather and shape are separate labelled steps | Action segments; set per program |
| Delivery format | How the data is packaged | RLDS, LeRobot, WebDataset, HDF5, zarr, Rerun .rrd or MCAP |
| Accept criteria | What passes QA, including what counts as a reject | Reject log per batch |
| Privacy | How people, plates and screens in frame are protected | Faces, plates and screens blurred at ingest |
| License | Who owns the dataset | Perpetual, buyer-owned commercial license; exclusivity available |
Example footage
One real clip already published on this site, shown as a format and rig example: ten seconds of first-person home cooking with faces blurred. It does not show glass etching or glassblowing.
No glass etching or glassblowing clip is published on the site, so this is the nearest first-person clip, labelled as a format example. Request a sample to see delivered files. See sample data outputs, request an egocentric data sample.
FAQ
Sources
Tell us the pieces and tools, whether an episode is etching, glassblowing or both, whether clear or reflective pieces are in scope, the views and streams you want and the format you train on. 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.