Egocentric Sanding and Surface Finishing Video Data

Egocentric sanding and surface finishing video data is first-person video of a person smoothing, polishing, buffing or deburring a surface, recorded from their 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 at a bench, so the frames show what the hands and the tool are doing to the surface rather than a distant view of a workshop. Teams searching for sanding robot training data, a surface finishing video dataset or egocentric polishing and sanding data are usually looking for this: footage of the work itself, tied to labels they can train on.

On this site the data is human capture: a paid, consented contributor wears the rig and performs a task from a written skill spec. It is not robot or teleoperation data, and it is not a fixed corpus you download. It is custom finishing task data collection, scoped to your tasks. This page is a different subject from woodworking and carpentry data, which covers measuring, cutting, joining and finishing wood as a whole, from painting and coating data, which centres on applying a material, from welding and metal fabrication data, and from wiping and surface cleaning data, where the aim is a clean surface rather than a finished one.

The task catalog names sanding, polishing, buffing and deburring rows in eight environments, fourteen rows in all. Three sit in the Carpentry & Furniture Workshop, whose example line reads "Joinery and furniture workshop with saws, sanders and finishing stations". One sands wood surfaces with an orbital sander through the grits and wipes on stain with rags. One sands panels smooth by machine and hand, fills defects and wipes off dust before finishing. One applies stain, sealer or lacquer and sands between coats. All three are rated Medium.

Two rows sit in the Metal Fabrication & Welding Workshop, grinding and deburring welded seams and cut edges, and two sit in the Plastic & Light-Assembly Manufacturer, trimming flash from molded parts and deburring their edges. The metal rows are rated Medium, the plastic rows Easy, and all four are Ready to deploy. A Trades Workshop Bench row, to sand a surface and wipe it clean, is rated Easy and Available on request. It is the one row recorded with a multiview rig and handheld grippers rather than a head-mounted phone.

Polishing and buffing appear in four more environments. Car Wash & Detailing lists applying, buffing and inspecting polishing compound on body panels, polishing exterior panels after a wash, and drying and buffing bodywork to finish, all Active collection. Auto Service Center & Tire Shop lists hand drying and buffing paintwork to a finish, Ready to deploy. Jewelry & Metalsmithing Studio lists polishing metal pieces on the wheel with compounds and finishing by hand, rated Medium. Repair Shop lists furniture restoration, which strips old finish, repairs joints and veneer, sands and refinishes the piece. It is the only Hard row, and with the jewelry row it is Available on request.

The first table below lists every row. Nail buffing, floor waxing and glass polishing are left out because they are beauty, floor-care and cleaning tasks, and wiping a surface is not finishing. The catalog states no grit value, tool model or part size for any row, so this page states none.

Each attempt is delivered as one episode. On this site the standard episode carries action segments, 3D hand pose for both hands with a per-joint visibility flag, contact events that mark when a hand meets an object, an objects list, and per-frame instance masks stored at the depth resolution of 848×480. The capture setup table below lists the streams and sync figures the site states for the instrumented tier: ten streams hardware-triggered onto one PTP clock domain and aligned under 2 ms, 6-DoF IMU data at 200 Hz, and calibration twice per session.

The catalog lists a head-mounted phone with a multiview rig for thirteen of these rows, and a multiview rig with handheld grippers for the trades bench row, while the ten-stream stack is the instrumented tier. Which applies to a finishing program is not stated and is confirmed in the skill spec. Force and tactile readings are not among the listed streams.

Action segments divide an episode into start- and end-timestamped spans, each carrying start_ns, end_ns, verb and noun. For finishing work, a segment might be one sanding pass, one change of abrasive, one wipe of the surface or one polishing round, as written in the skill spec. Where one completed step starts and ends is a spec decision, not a default.

Validation runs against signed accept criteria written into the spec before capture, and a reject log ships with each batch. Episodes are machine-checked against stated reject criteria, including a sync breach, calibration drift, the manipulated object out of frame for more than 15 percent of the episode, an undocumented bystander, and a mismatch against the named skill spec, and a human reviewer then checks each episode. Only footage that clears the criteria is billed as a validated hour, and rejected items are recaptured at our cost.

Delivery is in the format your stack already reads: RLDS, LeRobot, WebDataset, HDF5, zarr or Rerun .rrd, or raw streams with metadata, with readable converter source alongside. The license is perpetual, irrevocable and worldwide, buyer-owned and non-exclusive by default, with exclusivity available as a separate term. The executed agreement controls.

Finishing is repetition with a purpose. The catalog describes sanding "through the grits", sanding "between coats" and polishing then inspecting, so a model has to learn the order of passes and when one stops, not a single stroke. The tool sits against the surface and the fingers wrap it, so the contact point is often hidden, which is why hand pose carries a per-joint visibility flag, contact events mark when the hand meets the object and masks are recorded per frame. Whether a surface is finished is a judgment, and the site states no finish-quality label. The second table below maps each challenge to the stream or label this site states, and says so where nothing is stated.

A robot foundation model paper on arXiv organizes its training data as a pyramid, with large quantities of web data and human videos at the base, synthetic data in the middle and real robot data at the top. The same paper says human videos have no explicit action labels, so it learns latent actions to label them. That is robot-model research, not Firsthand data, and Firsthand has no relationship with its authors.

The catalog describes the head-mounted phone as recording the first-person view of real work on site. The site also states that Firsthand does not capture at a customer’s own facility. For the trades bench row, the site describes the handheld-gripper mode as a staged working set recorded while the operator uses the grippers. For the other rows, whether a program uses a working shop or a staged set is not stated, and neither is the venue type. Both are agreed in the skill spec before capture begins.

Every contributor signs a written release, and each delivered episode links to it through consent artefact IDs in the batch data card. The consent chain has six checkpoints, and a break at any one discards the footage. At ingest, automated detection finds faces, plates and screens and blurs them irreversibly. 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, footage that cannot be de-identified is discarded, and the anonymization record ships with the data.

The site states no safety, protective equipment, dust control or certification practice for these environments, so this page makes no such claim. A row that mentions wiping off dust describes a step the contributor performs, not a delivered measurement.

From signed spec to first validated batch, the published 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. Four rows are Active collection, where recording is running now and a spec can join an existing capture stream. Seven are Ready to deploy, with contributors and site access set up. Three are Available on request, scoped and recruited to spec. The range that applies is confirmed when the spec is quoted.

Describe the finishing tasks and their order, what counts as a finished pass or step, who may be in frame, and the variation cases your pipeline is missing. That becomes a written skill spec, quoted for reach, timeline and price before any capture begins. Our skill spec guide walks through how to write one.

Most programs begin with a Spec Pilot, a small, fast first batch that proves the spec and the accept criteria on real footage before volume scales. Collections are billed per validated hour. Counts of sites, contributors, episodes and hours are set in the skill spec, since the site does not state them for these environments.