Egocentric Data Collection Services for AI Training Data

On-demand egocentric data collection. We recruit participants, run first-person capture across 15+ countries, and deliver consented, annotated AI training data.(158 chars)

50,000+

Hours of Egocentric Video Collected

Real-world first-person recordings across multiple industries.

500+

Data Collection Projects

Successfully delivered datasets for AI and robotics applications.

15+

Countries Covered

Diverse participants, environments, and use cases for robust AI training.

verbosetechlabs vt icon verbosetechlabs vt icon Egocentric Data Collection Sample

Egocentric Data Collection Briefs We've Already Run

Real first-person collections for real teams, shown with the brief that started them. Open any card to see the spec we shot and what shipped.

First-person video sample: chopping vegetables (household egocentric dataset)
00:00

Chopping vegetables

Household

Training bimanual grasp and re-grasp policies on deformable objects. Captures the stabilising hand, mid-task re-grips, and natural recovery from slip.

  • VOLUME

    268 hrs

  • PARTICIPANTS

    24 operators

  • RESOLUTION + FPS

    1080p. 30 fps

  • CAMERA RIG

    Head-mounted

  • CAMERA

    iPhone 14 Pro

  • FORMAT

    MP4

  • SCALABLE TO

    1,000 hrs 8 weeks lead time

First-person video sample: cleaning the home (household egocentric dataset)
00:00

Cleaning the Home

Household

Training whole-body mobile manipulation across a lived-in space — reaching, wiping, and clearing surfaces amid natural clutter.

  • VOLUME

    268 hrs

  • PARTICIPANTS

    24 operators

  • RESOLUTION + FPS

    1080p. 30 fps

  • CAMERA RIG

    Head-mounted

  • CAMERA

    iPhone 14 Pro

  • FORMAT

    MP4

  • SCALABLE TO

    1,000 hrs 8 weeks lead time

First-person video sample: desk cleaning (household egocentric dataset)
00:00

Desk Cleaning

Household

Training tabletop tidying — grasping mixed small objects, wiping the surface, and returning items to place.

  • VOLUME

    268 hrs

  • PARTICIPANTS

    24 operators

  • RESOLUTION + FPS

    1080p. 30 fps

  • CAMERA RIG

    Head-mounted

  • CAMERA

    iPhone 14 Pro

  • FORMAT

    MP4

  • SCALABLE TO

    1,000 hrs 8 weeks lead time

First-person video sample: folding clothes (sample 1) (household egocentric dataset)
00:00

Folding Clothes (Sample 1)

Household

Training bimanual manipulation of highly deformable fabric — flattening, aligning edges, and forming repeatable folds.

  • VOLUME

    268 hrs

  • PARTICIPANTS

    24 operators

  • RESOLUTION + FPS

    1080p. 30 fps

  • CAMERA RIG

    Head-mounted

  • CAMERA

    iPhone 14 Pro

  • FORMAT

    MP4

  • SCALABLE TO

    1,000 hrs 8 weeks lead time

First-person video sample: making a bed (household egocentric dataset)
00:00

Making a Bed

Household

Training large-fabric handling over a wide workspace — spreading, smoothing, and tucking sheets that drape and buckle.

  • VOLUME

    268 hrs

  • PARTICIPANTS

    24 operators

  • RESOLUTION + FPS

    1080p. 30 fps

  • CAMERA RIG

    Head-mounted

  • CAMERA

    iPhone 14 Pro

  • FORMAT

    MP4

  • SCALABLE TO

    1,000 hrs 8 weeks lead time

First-person video sample: folding clothes (sample 2) (household egocentric dataset)
00:00

Folding Clothes (Sample 2)

Household

Training load-and-unload manipulation at an appliance — transferring limp garments into and out of a drum and operating controls.

  • VOLUME

    268 hrs

  • PARTICIPANTS

    24 operators

  • RESOLUTION + FPS

    1080p. 30 fps

  • CAMERA RIG

    Head-mounted

  • CAMERA

    iPhone 14 Pro

  • FORMAT

    MP4

  • SCALABLE TO

    1,000 hrs 8 weeks lead time

First-person video sample: washing clothes in a washing machine (household egocentric dataset)
00:00

Washing Clothes in a Washing Machine

Household

Training load-and-unload manipulation at an appliance — transferring limp garments into and out of a drum and operating controls.

  • VOLUME

    268 hrs

  • PARTICIPANTS

    24 operators

  • RESOLUTION + FPS

    1080p. 30 fps

  • CAMERA RIG

    Head-mounted

  • CAMERA

    iPhone 14 Pro

  • FORMAT

    MP4

  • SCALABLE TO

    1,000 hrs 8 weeks lead time

First-person video sample: cooking eggs (household egocentric dataset)
00:00

Cooking Eggs

Household

Training contact-rich cooking manipulation — cracking, pouring, and pan handling with a tool under changing material state.

  • VOLUME

    268 hrs

  • PARTICIPANTS

    24 operators

  • RESOLUTION + FPS

    1080p. 30 fps

  • CAMERA RIG

    Head-mounted

  • CAMERA

    iPhone 14 Pro

  • FORMAT

    MP4

  • SCALABLE TO

    1,000 hrs 8 weeks lead time

First-person video sample: cooking chapati (household egocentric dataset)
00:00

Cooking Chapati

Household

Training dexterous deformable-dough manipulation — rolling, shaping, flipping, and pan transfer of soft, sticky material.

  • VOLUME

    268 hrs

  • PARTICIPANTS

    24 operators

  • RESOLUTION + FPS

    1080p. 30 fps

  • CAMERA RIG

    Head-mounted

  • CAMERA

    iPhone 14 Pro

  • FORMAT

    MP4

  • SCALABLE TO

    1,000 hrs 8 weeks lead time

First-person video sample: making burgers at home (household egocentric dataset)
00:00

Making Burgers at Home

Household

Training repetitive bimanual food assembly — stacking, portioning, and ordered layering across many identical units.

  • VOLUME

    268 hrs

  • PARTICIPANTS

    24 operators

  • RESOLUTION + FPS

    1080p. 30 fps

  • CAMERA RIG

    Head-mounted

  • CAMERA

    iPhone 14 Pro

  • FORMAT

    MP4

  • SCALABLE TO

    1,000 hrs 8 weeks lead time

First-person video sample: cooking rice (household egocentric dataset)
00:00

Cooking Rice

Household

Training multi-step food prep — measuring, rinsing, pouring, and pot handling through a sequenced cooking process.

  • VOLUME

    268 hrs

  • PARTICIPANTS

    24 operators

  • RESOLUTION + FPS

    1080p. 30 fps

  • CAMERA RIG

    Head-mounted

  • CAMERA

    iPhone 14 Pro

  • FORMAT

    MP4

  • SCALABLE TO

    1,000 hrs 8 weeks lead time

First-person video sample: cooking food (household egocentric dataset)
00:00

Cooking Food

Household

Training long-horizon kitchen manipulation — combining tool-use, pouring, stirring, and heat-side handling across a full recipe.

  • VOLUME

    268 hrs

  • PARTICIPANTS

    24 operators

  • RESOLUTION + FPS

    1080p. 30 fps

  • CAMERA RIG

    Head-mounted

  • CAMERA

    iPhone 14 Pro

  • FORMAT

    MP4

  • SCALABLE TO

    1,000 hrs 8 weeks lead time

First-person video sample: washing clothes by hand (household egocentric dataset)
00:00

Washing Clothes by Hand

Household

Training wet deformable-fabric manipulation — scrubbing, wringing, and rinsing limp cloth under changing weight and grip.

  • VOLUME

    268 hrs

  • PARTICIPANTS

    24 operators

  • RESOLUTION + FPS

    1080p. 30 fps

  • CAMERA RIG

    Head-mounted

  • CAMERA

    iPhone 14 Pro

  • FORMAT

    MP4

  • SCALABLE TO

    1,000 hrs 8 weeks lead time

First-person video sample: ironing clothes (household egocentric dataset)
00:00

Ironing Clothes

Household

Training repetitive bimanual food assembly — stacking, portioning, and ordered layering across many identical units.

  • VOLUME

    268 hrs

  • PARTICIPANTS

    24 operators

  • RESOLUTION + FPS

    1080p. 30 fps

  • CAMERA RIG

    Head-mounted

  • CAMERA

    iPhone 14 Pro

  • FORMAT

    MP4

  • SCALABLE TO

    1,000 hrs 8 weeks lead time

First-person video sample: organising items in a room (household egocentric dataset)
00:00

Organising Items in a Room

Household

Training whole-body pick-and-place across a cluttered space — sorting mixed objects and returning them to defined locations.

  • VOLUME

    268 hrs

  • PARTICIPANTS

    24 operators

  • RESOLUTION + FPS

    1080p. 30 fps

  • CAMERA RIG

    Head-mounted

  • CAMERA

    iPhone 14 Pro

  • FORMAT

    MP4

  • SCALABLE TO

    1,000 hrs 8 weeks lead time

First-person video sample: cleaning beside shelf (household egocentric dataset)
00:00

Cleaning Beside Shelf

Household

Training constrained-space wiping and reaching — cleaning narrow, partially occluded areas around fixed furniture.

  • VOLUME

    268 hrs

  • PARTICIPANTS

    24 operators

  • RESOLUTION + FPS

    1080p. 30 fps

  • CAMERA RIG

    Head-mounted

  • CAMERA

    iPhone 14 Pro

  • FORMAT

    MP4

  • SCALABLE TO

    1,000 hrs 8 weeks lead time

First-person video sample: cleaning window (household egocentric dataset)
00:00

Cleaning Window

Household

Training vertical-surface tool-use — sustained wiping over a large flat plane with consistent coverage and pressure.

  • VOLUME

    268 hrs

  • PARTICIPANTS

    24 operators

  • RESOLUTION + FPS

    1080p. 30 fps

  • CAMERA RIG

    Head-mounted

  • CAMERA

    iPhone 14 Pro

  • FORMAT

    MP4

  • SCALABLE TO

    1,000 hrs 8 weeks lead time

First-person video sample: origami / sequence angle (household egocentric dataset)
00:00

Origami / sequence angle

Household

Training long-horizon deformable manipulation through multi-step paper folding. Captures ordered fold sequences, crease reinforcement, and mid-task re-alignment.

  • VOLUME

    268 hrs

  • PARTICIPANTS

    24 operators

  • RESOLUTION + FPS

    1080p. 30 fps

  • CAMERA RIG

    Head-mounted

  • CAMERA

    iPhone 14 Pro

  • FORMAT

    MP4

  • SCALABLE TO

    1,000 hrs 8 weeks lead time

First-person video sample: organising books in a bookshelf (household egocentric dataset)
00:00

Organising Books in a Bookshelf

Household

Training precise placement of rigid objects into tight slots — grasping, orienting, and inserting books against neighbours.

  • VOLUME

    268 hrs

  • PARTICIPANTS

    24 operators

  • RESOLUTION + FPS

    1080p. 30 fps

  • CAMERA RIG

    Head-mounted

  • CAMERA

    iPhone 14 Pro

  • FORMAT

    MP4

  • SCALABLE TO

    1,000 hrs 8 weeks lead time

First-person video sample: organising utensils in a rack (household egocentric dataset)
00:00

Organising Utensils in a Rack

Household

Training placement of varied rigid objects into structured holders — orientation and slotting under spatial constraint.

  • VOLUME

    268 hrs

  • PARTICIPANTS

    24 operators

  • RESOLUTION + FPS

    1080p. 30 fps

  • CAMERA RIG

    Head-mounted

  • CAMERA

    iPhone 14 Pro

  • FORMAT

    MP4

  • SCALABLE TO

    1,000 hrs 8 weeks lead time

First-person video sample: making a bed and arranging cushions (household egocentric dataset)
00:00

Making a Bed and Arranging Cushions

Household

Training bimanual assembly of soft goods — stuffing, shaping, and arranging compliant filled material.

  • VOLUME

    268 hrs

  • PARTICIPANTS

    24 operators

  • RESOLUTION + FPS

    1080p. 30 fps

  • CAMERA RIG

    Head-mounted

  • CAMERA

    iPhone 14 Pro

  • FORMAT

    MP4

  • SCALABLE TO

    1,000 hrs 8 weeks lead time

First-person video sample: watering plant (household egocentric dataset)
00:00

Watering Plant

Household

Training controlled pouring and reach — guiding liquid to targets while managing container weight and spill.

  • VOLUME

    268 hrs

  • PARTICIPANTS

    24 operators

  • RESOLUTION + FPS

    1080p. 30 fps

  • CAMERA RIG

    Head-mounted

  • CAMERA

    iPhone 14 Pro

  • FORMAT

    MP4

  • SCALABLE TO

    1,000 hrs 8 weeks lead time

First-person video sample: washing utensils (sample 1) (household egocentric dataset)
00:00

Washing Utensils (Sample 1)

Household

Training wet-object bimanual manipulation — grasping slippery items, scrubbing with a tool, rinsing, and controlled placement.

  • VOLUME

    268 hrs

  • PARTICIPANTS

    24 operators

  • RESOLUTION + FPS

    1080p. 30 fps

  • CAMERA RIG

    Head-mounted

  • CAMERA

    iPhone 14 Pro

  • FORMAT

    MP4

  • SCALABLE TO

    1,000 hrs 8 weeks lead time

First-person video sample: washing utensils (sample 2) (household egocentric dataset)
00:00

Washing Utensils (Sample 2)

Household

Training wet-object bimanual manipulation — grasping slippery items, scrubbing with a tool, rinsing, and controlled placement.

  • VOLUME

    268 hrs

  • PARTICIPANTS

    24 operators

  • RESOLUTION + FPS

    1080p. 30 fps

  • CAMERA RIG

    Head-mounted

  • CAMERA

    iPhone 14 Pro

  • FORMAT

    MP4

  • SCALABLE TO

    1,000 hrs 8 weeks lead time

First-person video sample: washing utensils (sample 3) (household egocentric dataset)
00:00

Washing Utensils (Sample 3)

Household

Training wet-object bimanual manipulation — grasping slippery items, scrubbing with a tool, rinsing, and controlled placement.

  • VOLUME

    268 hrs

  • PARTICIPANTS

    24 operators

  • RESOLUTION + FPS

    1080p. 30 fps

  • CAMERA RIG

    Head-mounted

  • CAMERA

    iPhone 14 Pro

  • FORMAT

    MP4

  • SCALABLE TO

    1,000 hrs 8 weeks lead time

First-person video sample: cutting metal with an angle grinder (fabrication egocentric dataset)
00:00

Cutting metal with an angle grinder

Fabrication

Training two-hand operation of a high-force powered tool against rigid material. Captures the firm dual-grip, controlled feed into the cut, and steady reaction-force management as the grinder bites — high-load tool-use that free-space manipulation data can't provide.

  • VOLUME

    268 hrs

  • PARTICIPANTS

    24 operators

  • RESOLUTION + FPS

    1080p. 30 fps

  • CAMERA RIG

    Head-mounted

  • CAMERA

    iPhone 14 Pro

  • FORMAT

    MP4

  • SCALABLE TO

    1,000 hrs 8 weeks lead time

First-person video sample: metal welding (fabrication egocentric dataset)
00:00

Metal Welding

Fabrication

Training precise tool guidance under an active high-energy process. Captures the hand striking and maintaining an arc — controlled electrode angle, standoff distance, and steady travel along the joint — fine positioning under extreme conditions that no simulated data reproduces.

  • VOLUME

    268 hrs

  • PARTICIPANTS

    24 operators

  • RESOLUTION + FPS

    1080p. 30 fps

  • CAMERA RIG

    Head-mounted

  • CAMERA

    iPhone 14 Pro

  • FORMAT

    MP4

  • SCALABLE TO

    1,000 hrs 8 weeks lead time

First-person video sample: cutting grass with a sickle (agriculture egocentric dataset)
00:00

Cutting grass with a sickle

Agriculture

Training bimanual tool-use on compliant vegetation in unstructured terrain. Captures the coordinated split of roles — one hand bunching and tensioning grass, the other drawing a curved blade through it — under variable resistance that clean tabletop tool data can't reproduce.

  • VOLUME

    268 hrs

  • PARTICIPANTS

    24 operators

  • RESOLUTION + FPS

    1080p. 30 fps

  • CAMERA RIG

    Head-mounted

  • CAMERA

    iPhone 14 Pro

  • FORMAT

    MP4

  • SCALABLE TO

    1,000 hrs 8 weeks lead time

First-person video sample: harvesting chillies (agriculture egocentric dataset)
00:00

Harvesting chillies

Agriculture

Training bimanual selective picking on live plants in unstructured field conditions. Captures one hand steadying the stem while the other locates and detaches ripe produce without damaging the plant — delicate targeted harvesting that clean tabletop data can't reproduce

  • VOLUME

    268 hrs

  • PARTICIPANTS

    24 operators

  • RESOLUTION + FPS

    1080p. 30 fps

  • CAMERA RIG

    Head-mounted

  • CAMERA

    iPhone 14 Pro

  • FORMAT

    MP4

  • SCALABLE TO

    1,000 hrs 8 weeks lead time

First-person video sample: soldering a circuit board (industrial egocentric dataset)
00:00

Soldering a circuit board

Industrial

Training precision bimanual tool-use for electronics assembly. Captures the steadying hand fixturing the board while the tool hand places the iron with sub-millimetre accuracy — controlled contact, dwell, and withdrawal that vision-only policies can't learn.

  • VOLUME

    268 hrs

  • PARTICIPANTS

    24 operators

  • RESOLUTION + FPS

    1080p. 30 fps

  • CAMERA RIG

    Head-mounted

  • CAMERA

    iPhone 14 Pro

  • FORMAT

    MP4

  • SCALABLE TO

    1,000 hrs 8 weeks lead time

First-person video sample: handling sheet-metal parts (industrial egocentric dataset)
00:00

Handling sheet-metal parts

Industrial

Training whole-body bimanual handling of large rigid parts in a working facility. Captures lifting, orienting, and placing heavy sheet-metal panels — including two-person coordinated moves — behaviours that tabletop manipulation data can't provide.

  • VOLUME

    268 hrs

  • PARTICIPANTS

    24 operators

  • RESOLUTION + FPS

    1080p. 30 fps

  • CAMERA RIG

    Head-mounted

  • CAMERA

    iPhone 14 Pro

  • FORMAT

    MP4

  • SCALABLE TO

    1,000 hrs 8 weeks lead time

First-person video sample: screen printing (industrial egocentric dataset)
00:00

Screen printing

Industrial

Training bimanual handling of large flexible-framed tools in a production setting. Captures lifting, positioning, and registering a screen-print frame — two-hand control of an awkward, semi-rigid object against a fixed print bed that tabletop data can't reproduce.

  • VOLUME

    268 hrs

  • PARTICIPANTS

    24 operators

  • RESOLUTION + FPS

    1080p. 30 fps

  • CAMERA RIG

    Head-mounted

  • CAMERA

    iPhone 14 Pro

  • FORMAT

    MP4

  • SCALABLE TO

    1,000 hrs 8 weeks lead time

First-person video sample: operating a power press (industrial egocentric dataset)
00:00

Operating a power press

Industrial

Training bimanual guidance of deformable fabric through a powered machine. Captures both hands feeding, aligning, and tensioning cloth under the needle in real time — continuous fine control of limp material that tabletop pick-and-place data can't provide.

  • VOLUME

    268 hrs

  • PARTICIPANTS

    24 operators

  • RESOLUTION + FPS

    1080p. 30 fps

  • CAMERA RIG

    Head-mounted

  • CAMERA

    iPhone 14 Pro

  • FORMAT

    MP4

  • SCALABLE TO

    1,000 hrs 8 weeks lead time

First-person video sample: ironing fabric (industrial egocentric dataset)
00:00

Ironing fabric

Industrial

Training bimanual pressing of deformable material with a handheld tool. Captures one hand smoothing and positioning cloth while the other guides the iron with controlled pressure and path — coordinated flatten-and-press behaviour on limp fabric that rigid manipulation data lacks.

  • VOLUME

    268 hrs

  • PARTICIPANTS

    24 operators

  • RESOLUTION + FPS

    1080p. 30 fps

  • CAMERA RIG

    Head-mounted

  • CAMERA

    iPhone 14 Pro

  • FORMAT

    MP4

  • SCALABLE TO

    1,000 hrs 8 weeks lead time

First-person video sample: trimming garment threads (industrial egocentric dataset)
00:00

Trimming garment threads

Industrial

Training fine bimanual manipulation of soft fabric with a small tool. Captures both hands working in close coordination — one holding and tensioning cloth, the other trimming stray threads — precise close-range work on deformable material that rigid-object policies never encounter.

  • VOLUME

    268 hrs

  • PARTICIPANTS

    24 operators

  • RESOLUTION + FPS

    1080p. 30 fps

  • CAMERA RIG

    Head-mounted

  • CAMERA

    iPhone 14 Pro

  • FORMAT

    MP4

  • SCALABLE TO

    1,000 hrs 8 weeks lead time

First-person video sample: packing metal parts (industrial egocentric dataset)
00:00

Packing metal parts

Industrial

Training bimanual pick-and-pack from bulk clutter into flexible packaging. Captures reaching into a mixed bin, grasping small rigid parts, and inserting them into limp plastic bags — the combined rigid-grasp plus deformable-bag handling that real packing lines demand.

  • VOLUME

    268 hrs

  • PARTICIPANTS

    24 operators

  • RESOLUTION + FPS

    1080p. 30 fps

  • CAMERA RIG

    Head-mounted

  • CAMERA

    iPhone 14 Pro

  • FORMAT

    MP4

  • SCALABLE TO

    1,000 hrs 8 weeks lead time

verbosetechlabs vt icon verbosetechlabs vt icon Rights and handling

Egocentric Data You Can Actually Train On

The cheapest dataset is the one you can't use. Every egocentric data collection project ships with the paperwork that lets your legal team say yes.

 Signed participant consent  Signed participant consent

Signed participant consent

Commercial training rights granted explicitly, per participant, on file.

GDPR-aligned handling GDPR-aligned handling

GDPR-aligned handling

Face blurring, voice masking, and PII redaction applied before delivery on request on demand.

ISO 20000 certified process ISO 20000 certified process

ISO 20000 certified process

Secure transfer, access logs, NDAs signed before the first frame is shot.

 Exclusive or shared licence  Exclusive or shared licence

Exclusive or shared licence

Keep the dataset yours, or price it down by letting it enter the OTS library later.

verbosetechlabs vt icon Get Started Today

Tell us what to record. We'll tell you what it takes.

Send the brief you built above, or just describe the failure case your model keeps hitting. A data lead — not a sales rep — replies within 24 hours with an egocentric data collection spec and a fixed quote.

Tell us what to record. We'll tell you what it takes.

Frequently Asked Questions About AI Development Company Services

Answers to help you choose the right AI development services, ready-made apps, and scalable solutions for your business.

On-demand data collection is the process of capturing new, purpose-built training data for a specific AI model rather than licensing an existing dataset. You define the activity, environment, participant profile, and volume; the vendor recruits people, deploys the recording rigs, runs capture, and delivers annotated data with consent and licensing on file. Teams choose it when the footage their model needs doesn't exist in any public or off-the-shelf dataset.

Off-the-shelf datasets are already recorded and ready to license today. Custom collection is recorded to your spec and typically takes two to three weeks before the first batch arrives. Off-the-shelf is faster and cheaper; custom gives you the exact activities, edge cases, demographics, and camera angles your model is failing on — and usually an exclusive licence. Most teams start with samples to validate quality, then commission a custom run.

The first batch usually arrives in two to three weeks. Scoping takes two to three days, participant recruitment and consent one to two weeks, and a pilot capture three to five days. Full capture then scales with volume — a 500-hour egocentric video collection typically completes in six to eight weeks, while 2,000+ hours run in parallel across multiple sites and countries.

Cost is quoted per hour of usable, QA-passed data and depends on four variables: the complexity of the activity, the environment (a hospital ward costs more to access than a home kitchen), the sensor stack, and the annotation depth. Raw footage is cheapest; segmentation and pose labelling sit at the top. We quote a fixed per-hour rate at the scoping stage so volume doesn't produce surprises later.

Egocentric video is footage recorded from a person's own viewpoint using a head-, chest-, or wrist-mounted camera, so the hands and the manipulated objects stay in frame. It teaches models hand-object interaction, task sequencing, and intent — signals that fixed cameras miss. It's the primary training modality for robot manipulation policies, embodied AI agents, and assistive smart-glasses applications.