Egocentric Tactile Glove and Hand Tracking Video Data

Egocentric tactile glove data pairs a first-person camera view with a signal recorded by gloves the contributor wears while doing real work. The gloves track the hands, and the glove signal is time-synced to a head-mounted stereo camera by hardware timestamps and an IMU. It is the tier above camera-only capture.

The hand tracking page covers pose from cameras, and the contact event page covers a label for when contact changes. Neither describes a worn sensor. This page covers the glove tier, what is stated about it, and what is left to the written spec.

Standard Firsthand capture uses cameras, depth and IMU streams. Contact events in that capture are labels that say when and where contact changed, and they are not described as force, pressure or tactile data. The glove tier adds gloves that the contributor wears, and with them a glove signal that sits beside the video.

The two tiers are not interchangeable. If your policy only needs to know when a hand touched an object, camera-only capture with contact events may be enough. If it needs a signal from the hand itself, the glove tier is the one to specify.

The glove tier prioritises dexterous tasks: assembly and kitting, packing, electronics repair, tool use, food preparation and skilled trades. These are the settings where finger placement and grip change from one step to the next.

The site has task pages for several of them, including tool use, electronics assembly and dexterous manipulation. Those pages describe the tasks and camera-based capture. Specify gloves on top of them when your policy needs the extra signal.

The glove kit is time-synced to a head-mounted stereo camera through hardware timestamps and an IMU. The kit is sourced and configured by Firsthand, so the sync setup is the same for every contributor in a program.

The site states a cross-stream sync figure of under 2 ms for its standard ten-stream episode. It does not state a sync error figure for the glove tier, so that number is Not stated and should be requested as part of the pilot.

Each contributor gets a glove calibration at the start of every session. Glove contributors also get an extra fitting and calibration session before they start, because a glove that fits one hand loosely will not read like one that fits another.

Kits are charged and checked by an on-site tech before capture. The site does not state a calibration method, a calibration target or a calibration error figure for gloves, so those are Not stated.

The glove tier uses the same upload, QA and delivery flow as camera-only capture, plus glove signal checks for dropout and sync drift. Dropout means the signal stops for a stretch. Sync drift means it slowly slides away from the video clock.

Each batch is validated against signed accept criteria before it ships, with a reject log. Pass thresholds for the glove checks are Not stated, so set them in the spec and agree them in the pilot.

Gloves with hand tracking are one route to hand pose. Two others are available today: vision-based 3D hand-pose annotation on stereo footage, and a stereo headset with built-in hand tracking. A spec can name one route or compare them on the same task.

The site states 3D hand pose with per-joint visibility for its standard capture. It does not state a joint count, a pose accuracy figure or a glove tracking rate, so those are Not stated.

Glove contributors come from the same trained contributor pool as camera-only capture, with the extra fitting and calibration session. An on-site tech charges and checks the kits, so a glove that has lost power or signal is caught before a session starts.

The site does not state the number of glove contributors, kits or sites, so those figures are Not stated. Capacity for the tier is quoted against your spec.

The glove tier is in pilot in the United States and Brazil. It uses the same working environments as camera-only capture, with non-residential settings first and residential settings available. Other countries are Not stated for this tier and are confirmed per engagement.

The site does not state capture at a customer own facility, so environment, access and schedule are agreed in the spec. Cities, studios and site counts are Not stated.

A glove episode carries the first-person stereo camera view, the hand tracking from the gloves and the glove signal on one timeline. The table below sets each against what the site states for camera-only capture.

Depth, wrist views and the ten-stream layout are stated for the standard episode. They are Not stated for the glove tier, so list the streams you need in the spec.

Labels are set by your spec. The site offers frame-accurate action segments and 3D hand pose on its standard capture, and vision-based 3D hand-pose annotation on stereo footage. Which of them apply to a glove episode is agreed with the tasks.

Force or pressure units, tactile event labels and grasp-quality labels are Not stated. If your policy needs a specific tactile label, define it in the spec so the pilot can test it.

Accepted data is packaged in your format, such as MP4 with JSON or CSV, MCAP, LeRobot or a custom layout. The site also states RLDS, WebDataset, HDF5, zarr and Rerun .rrd. The format table below sets each against what its official documentation states.

How the glove signal maps into each format is Not stated and is set in the spec. A container that stores timestamped messages, such as MCAP, can hold a continuous signal beside the video.

Every episode links to a signed contributor release through a documented consent chain, and the consent record is part of the delivery. Faces, plates and screens are detected and irreversibly blurred at ingest, and a reviewer confirms the result before anything ships.

The site states no privacy rule specific to gloves, so this page makes none. The consent chain, the blur pass and the anonymization record are the stated steps.

The glove tier is pilot-ready, and a timeline for it is quoted against your spec before capture begins. The site does not state a delivery time for glove data, so none is promised here.

The 48-hour median and three-to-four-week figures that the site states apply to camera-based collection, where coverage is deep or thin. On-demand top-ups run against the same spec and accept criteria once a pilot has set them.

You describe the dexterous tasks and what your policy needs from the hands. We turn that into a written spec covering tasks, environments, glove and camera setup, labels and accept criteria, then quote reach, timeline and price before capture begins.

A Spec Pilot, a small first batch, can test the glove signal checks and the sync before you scale. Billing is per validated hour, a reject log ships with every batch, rejected items are recaptured at our cost, and every collection ships under a perpetual, buyer-owned commercial license.