End-to-end custom data collection for AI.

1. 01DiscoveryWe start from the failure, not a schema. You describe what you are training and where it breaks.
2. 02SpecificationThe problem becomes a written, testable spec you approve before anyone is briefed.
3. 03Sourcing & staffingWe recruit and screen the exact contributors the spec calls for, wherever they are.
4. 04CaptureContributors are briefed, paid hourly, and collected against the spec — with the right rig where it matters.
5. 05Annotation & structuringRaw capture is labelled, transcribed, or structured into the schema your loader expects.
6. 06QA & anonymizationAutomated checks plus human review against the accept criteria, with a visible reject log.
7. 07Licensed deliveryData ships in your format with the consent artefacts and a buyer-owned commercial license.

Each stage broken down — what happens, what you get, and how it is measured — on the end-to-end pipeline page. Want the whole thing run for you? See custom data collection as a service. For first-person everyday routines at weekly volume, see Insta360 head-mount routine collection. Planning a brief? Read how to scope a custom robotics data collection project.

The full breakdown by model type, with what we collect and an example brief for each, is on custom data collection for AI training.

• Physical AIPhysical AI training data →Physical AI training data is the sensor and video data used to train models that act in the physical world — robots, humanoids, and other embodied agents.
• Robotics video dataVideo data collection for robotics →Video data collection for robotics is recording task video from the viewpoint and sensors a robot policy will actually use, rather than generic third-person footage.
• On demandOn-demand data collection →On-demand data collection is custom AI training data gathered when you need it, against a spec you set, from contributors who are already recruited, briefed, and paid through a standing pipeline.
• Fast turnaroundFast data collection →With Firsthand, the first validated batch lands in a median of 48 hours from a signed spec in environments where coverage is already deep.
• Stereoscopic captureStereoscopic video data collection →Stereoscopic video data collection is capturing the same scene through two calibrated cameras spaced like human eyes, so every frame pair carries recoverable depth.
• Hard environmentsEgocentric stereo capture in hard environments →It is first-person video recorded with a calibrated stereo pair at human-eye spacing in demanding workplaces, such as appliance, hardware, electronics, lighting and furniture plants, so every frame pair carries recoverable depth.
• Appliance and metering plantsEgocentric stereo capture for appliance and metering plants →It is first-person video from a calibrated stereo pair, hardware-synced with depth, IMU and hand pose, recorded for work typical of air conditioner, gas appliance, refrigerator and water meter plants, and collected custom to a written skill spec.
• Hardware and light-assembly plantsEgocentric stereo capture for hardware and light-assembly plants →It is first-person video from a calibrated stereo pair, hardware-synced with depth, IMU and hand pose, recorded for work typical of door and window hardware, electronics assembly, lighting products and furniture plants, and collected custom to a written skill spec.
• Multi-camera captureMultiview video data collection →Multiview video data collection is recording a task through more than one camera at once — typically a head-mounted view for scene context plus wrist-mounted views for grasp-level detail — hardware-triggered onto one clock and calibrated against each other, rather than several unsynchronized recordings aligned afterward.
• Head-mounted captureHead-mounted camera data collection →Head-mounted camera data collection is recording video and synchronized sensor streams from a camera worn on the head, capturing a task from roughly the actor’s own eye line rather than from across the room.
• Global-shutter captureGlobal shutter egocentric video data →Global shutter egocentric video data is first-person footage recorded by a camera that exposes every pixel in the frame at the same instant, so fast head and hand motion does not skew or wobble the image.
• Synchronized captureTime-synchronized egocentric sensor data →Time-synchronized egocentric sensor data is first-person task footage recorded together with the other sensor streams of the same moment, every stream stamped against one shared clock.
• Embodied AIEmbodied AI data collection →Embodied AI data collection is recording people performing physical tasks from their own point of view, with synchronized video, depth, motion, and hand pose, so models that act in the world can learn manipulation, navigation, and task structure from human demonstration.
• Egocentric videoEgocentric video data collection →Egocentric video data collection is recording video from the point of view of the person doing a task, using a head, chest, or glasses-mounted camera.
• Commercial licensingCommercially licensed egocentric data →Commercially licensed egocentric data is first-person video captured specifically for a buyer, with every contributor’s consent documented and a commercial license grant attached, so the buyer can train and ship a model on it without relying on a research-only public dataset or unclear provenance.
• Exclusive dataExclusive egocentric training data →Exclusive egocentric training data is custom first-person video collected to one buyer’s skill spec, where the contract restricts who else may be licensed the same footage or, depending on the negotiated scope, data captured to the same spec.
• First-party dataFirst-party training data →First-party training data is data collected directly by the provider that delivers it, from contributors it recruits, briefs, pays, and holds a signed release for, rather than scraped from the web or licensed onward from someone else’s collection.
• First-party egocentric dataFirst-party egocentric video data →First-party egocentric video data is first-person footage recorded by contributors the provider recruits, briefs, pays, and holds a release for, captured against a buyer’s spec rather than collected from public video or third-party archives.
• Our contributor networkContributor network data collection →Contributor network data collection is the gathering of training data by a provider’s own network of recruited, briefed contributors who capture against a written spec, rather than by scraping or licensing existing material.
• Privacy-safe captureAnonymized egocentric video data →Anonymized egocentric video data is first-person video captured to a spec where un-consented people, licence plates, and legible screens or documents are detected and irreversibly blurred before delivery, and spoken identifiers are scrubbed from audio.
• Delivery formatCustom LeRobot dataset →A custom LeRobot dataset is an egocentric video collection, captured to a buyer’s written spec, that is delivered in the LeRobot format: parquet metadata plus one mp4 per camera, laid out the way the Hugging Face datasets ecosystem and the LeRobot imitation-learning stack expect.
• Delivery formatCustom RLDS dataset →A custom RLDS dataset is an egocentric video collection, captured to a buyer’s written skill spec, that is delivered in the RLDS layout: TFDS-compatible tfrecord shards in which each episode is a sequence of steps carrying observations and an action.
• Delivery formatCustom HDF5 robot dataset →A custom HDF5 robot dataset is an egocentric video collection, captured to a buyer’s written skill spec, that is delivered as HDF5: one group per episode holding the synchronized streams, the annotations and the calibration, with an int64 nanosecond timestamp for every stream.
• Delivery formatCustom WebDataset video dataset →A custom WebDataset video dataset is an egocentric video collection, captured to a buyer’s written skill spec, that is delivered as WebDataset: .tar shards of 512 samples, each sample holding head and wrist images, a 16-bit depth image and an ann.json of annotations, streamed from signed URLs.
• Delivery formatCustom egocentric dataset in Rerun format →A custom egocentric dataset in Rerun format is first-person video collected to a buyer’s written skill spec and delivered as Rerun .rrd recordings: one recording per episode, with every stream on one shared nanosecond timeline.
• Delivery formatEgocentric stereo data in MCAP →Egocentric stereo data in MCAP format is first-person video from two calibrated cameras, with hand pose and sensor streams, delivered as .mcap files.
• Construction crewsStereo egocentric data from construction crews →Stereo egocentric data from construction crews is first-person video from two calibrated cameras, recorded from a crew member’s own viewpoint while they do real jobsite work, with hand pose and sensor streams alongside.
• Consent-backed captureConsented egocentric video data →Consented egocentric video data is first-person footage captured to a buyer’s spec where every contributor signed a written release before capture, identity was verified, and each delivered episode can be traced back to that consent record.
• Planned diversityDiverse egocentric video data →Diverse egocentric video data is first-person footage whose spread across contributors, scenes, tasks, objects, conditions and failure cases was planned in a written skill spec, with a minimum coverage target per axis that can be checked against the delivered batch.
• Real-home captureIn-home egocentric data collection →In-home egocentric data collection is the custom capture of first-person video inside real homes by paid, consented contributors wearing a capture rig, doing household tasks named in a written skill spec.
• Research labsEgocentric data for robotics research labs →Egocentric data for robotics research labs is first-person video collected against a lab’s own written skill spec rather than taken from a public dataset.
• First-person videoFirst-person video data collection →First-person video data collection is recording video from the actor’s own point of view, using a head, chest, or glasses-mounted camera, rather than filming them from across the room.
• ServicesEgocentric data collection services →An egocentric data collection service is a managed program that captures first-person human video and sensor data to a written spec, rather than selling a fixed, already-captured dataset.
• Robotics, on demandOn-demand robotics training data →On-demand robotics training data is human demonstration data collected when a robot policy needs it, scoped to the specific tasks, objects, or environments where it fails, and delivered labelled for imitation learning rather than bought as a fixed off-the-shelf dataset.
• Pretraining dataPretraining data collection for robot foundation models →Pretraining data collection for robot foundation models is the large-scale, general-purpose capture of first-person human behavior — video, depth, hand pose, and action labels — used to train the base layer of a robot foundation model before it is aligned to any specific robot or task.
• Pretraining dataEgocentric pretraining data →Egocentric pretraining data is first-person human video — captured from a head, chest, or glasses-mounted camera — used at scale to pretrain the general base layer of a robot foundation model.
• Action labelsAction-labeled egocentric video →Action-labeled egocentric video is first-person footage paired with structured labels for what the hands did — 3D hand pose, wrist trajectories, contact and grasp events, and start/end-timestamped action segments — rather than video alone or a single caption per clip.
• Frontier AI labsCustom data capture for frontier AI labs →Custom data capture for frontier AI labs is real-world, first-person training data collected against a lab's own written spec instead of licensed off the shelf.
• Humanoid robotsCustom training data for humanoid robots →Custom training data for humanoid robots is real-world, first-person human demonstration data collected against a written spec that matches a humanoid embodiment: a head camera viewpoint, two hands in frame, and whole-body motion.
• Humanoid VLAHumanoid VLA training data →Humanoid VLA training data is the demonstration data a vision-language-action model for a humanoid learns from: camera observations, a task instruction and an action signal.
• ManipulationCustom bimanual manipulation data collection →Custom bimanual manipulation data collection is real-world, first-person demonstration data of two-handed tasks captured against a written spec: egocentric video with per-hand 3D pose, contact events, and depth for both hands, aligned on one clock.
• Pretraining dataDexterous hand pretraining data →Dexterous hand pretraining data is large-scale, first-person capture of human hands performing varied manipulation tasks, with 3D hand pose, contact events, and wrist trajectories synchronized to video and depth.
• ManipulationDexterous manipulation training data →Dexterous manipulation training data is real-world, first-person demonstration data of fine-grained, multi-finger manipulation captured against a written spec: egocentric video with 3D hand pose, contact events, and depth.
• Warehouse & logisticsWarehouse robot training data →Warehouse robotics training data is real-world, first-person demonstration data of pick, scan, pack and pallet-build tasks captured against a written spec: egocentric video with depth, hand pose, and action segments.
• Task familyGrocery bagging robot training data →Egocentric grocery bagging and checkout task data is real-world, first-person video of a person scanning items, taking payment and bagging goods for a customer, captured with 3D hand pose, depth, contact events and frame-accurate action segments on a written spec.
• Task familySorting task robot training data →Egocentric sorting task data is real-world, first-person video of a person assigning items to destinations, such as returns bins, trays, carts or lanes, captured with 3D hand pose, depth, contact events and frame-accurate action segments on a written spec.
• Task familyBin picking training data →Egocentric bin picking data is real-world, first-person video of a person picking individual items from bins and placing them into totes, captured with 3D hand pose, depth, contact events and frame-accurate action segments on a written spec.
• Task familyVisual inspection training data →Egocentric visual inspection task data is real-world, first-person video of a person inspecting items: looking at a part, handling and rotating it, checking it against a standard and deciding whether to keep, set aside or reject it.
• Task familyPouring and liquid handling training data →Egocentric pouring and liquid handling data is real-world, first-person video of a person moving liquid from one vessel to another: pouring a drink, transferring between containers, measuring to a mark, or drawing and dispensing with a pipette.
• Task familyUnboxing and unpacking training data →Egocentric unboxing and unpacking data is real-world, first-person video of a person getting into a container and taking out what is inside: opening a carton, bag or pack, removing inserts and packaging, lifting the items out, checking them and setting them down.
• Task familyContainer opening robot training data →Egocentric container opening data is real-world, first-person video of a person opening a closed container: unscrewing a jar, opening a bottle, working a can open with a can opener, or capping and uncapping a sample tube.
• Task familyPacking and kitting training data →Egocentric packing and kitting data is real-world, first-person video of a person packing cartons, bags and containers, sealing and labeling them, or building kits from parts, captured with 3D hand pose, depth and frame-accurate action segments on a written spec.
• Task familyParcel handling robot training data →Egocentric box and parcel handling video data is real-world, first-person video of a person stacking, carrying, palletizing, sealing, scanning and sorting boxes and parcels, captured with 3D hand pose, depth and frame-accurate action segments on a written spec.
• Task familyTote induction robot training data →Egocentric tote induction and conveyor loading data is real-world, first-person video of a person inducting totes onto a conveyor, loading items onto a moving conveyor, nesting and stacking empty totes and picking items to a tote, captured with 3D hand pose, depth and frame-accurate action segments on a written spec.
• Task familyTruck unloading robot training data →Egocentric truck loading and unloading video data is real-world, first-person video of a person staging pallets and parcels at a dock, loading them into a trailer in sequence and unloading inbound goods from a truck, captured with 3D hand pose, depth and frame-accurate action segments on a written spec.
• Task familyWatch repair video data →Egocentric watch and small mechanism repair data is real-world, first-person video of a person opening a watch case, replacing the battery or movement parts under magnification and resealing it, captured with 3D hand pose, depth and frame-accurate action segments on a written spec.
• Task familyFurniture restoration video data →Egocentric furniture restoration video data is real-world, first-person video of a person stripping an old finish, repairing joints and veneer, then sanding and refinishing a piece of furniture, captured with 3D hand pose, depth and frame-accurate action segments on a written spec.
• Task familyFurniture restoration robot training data →Stereo furniture restoration and refinishing data is first-person video from two calibrated cameras of a person stripping an old finish, repairing joints and veneer, then sanding and refinishing a piece of furniture, with depth and hand pose alongside.
• Task familyAssisted living robot training data →Egocentric elder care and assisted living task data is real-world, first-person video of care-support tasks such as feeding assistance, medication handling, personal hygiene care, resident mobility assistance and room preparation, captured with 3D hand pose, depth and frame-accurate action segments on a written spec.
• Care tasksCare robot mobility assistance training data →It is custom first-person video of paid, consented contributors performing mobility assistance tasks such as steadying a mobility aid, supporting a person to stand, transfer or walk, and guiding balance or stretching work, captured to a written spec and delivered with synchronized streams, hand pose, contact events and action segments, anonymization records and a commercial training license.
• Task familyHospital robot training data →Egocentric hospital and clinical care task data is first-person video of hospital tasks such as equipment setup, specimen handling, patient room prep, surface decontamination and procedures on a phantom or mannequin, captured with 3D hand pose, depth and frame-accurate action segments on a written spec.
• Task familyClinical equipment setup robot training data →Egocentric clinical equipment setup data is first-person video of a person positioning and connecting monitors, pumps and stands at the bedside and checking they run, captured with 3D hand pose, depth and frame-accurate action segments on a written spec.
• Task familyLobby cleaning robot training data →Egocentric hotel lobby and common area cleaning data is first-person video of a person vacuuming and mopping lobby floors, dusting furniture, polishing glass and emptying bins, captured with 3D hand pose, depth and frame-accurate action segments on a written spec.
• Task familyClinical skills robot training data →Egocentric first aid and clinical skills training data is first-person video of practice tasks on a training mannequin or phantom, such as bandaging, CPR, applying a BP cuff, changing a dressing, setting up a sterile field and passing instruments, captured with 3D hand pose, depth and frame-accurate action segments on a written spec.
• Task familyTrash handling robot training data →Egocentric trash and recycling task video data is real-world, first-person video of a person pulling full bags, bagging trash and recycling, relining cans, emptying bins and carrying bags to the dumpsters, captured with 3D hand pose, depth and frame-accurate action segments on a written spec.
• Task familyCoffee robot training data →Egocentric beverage and coffee prep video data is real-world, first-person video of a person making drinks: grinding beans, dosing and tamping, pulling espresso, steaming milk, pouring, garnishing and operating a coffee machine, captured with 3D hand pose, depth and frame-accurate action segments on a written spec.
• Task familyGrocery put-away robot training data →Egocentric grocery put-away video data is real-world, first-person video of a person putting groceries away: taking items out of bags, carrying them, opening a cupboard and placing each item in storage, captured with 3D hand pose, depth and frame-accurate action segments on a written spec.
• Task familyTire change robot training data →Egocentric tire and wheel service video data is real-world, first-person video of a person servicing tires and wheels: dismounting and mounting a tire, balancing a wheel, rotating wheels between positions and checking pressure, captured with 3D hand pose, depth and frame-accurate action segments on a written spec.
• Task familyBrake service robot training data →Egocentric brake pad and rotor service data is real-world, first-person video of a person replacing brake pads and rotors: jacking the car, removing the wheel, compressing the caliper, fitting new pads and a rotor and torquing the bolts, captured with 3D hand pose, depth and frame-accurate action segments on a written spec.
• Task familyCar wash robot training data →Egocentric car washing and detailing video data is real-world, first-person video of a person washing, rinsing, drying, polishing and vacuuming a vehicle, captured with 3D hand pose, depth and frame-accurate action segments on a written spec.
• Task familyWelding robot training data →Egocentric welding and metal fabrication video data is real-world, first-person video of a person laying out, cutting, fitting, welding, grinding and inspecting metal parts, captured with 3D hand pose, depth and frame-accurate action segments on a written spec.
• Task familyKnifemaking robot training data →Egocentric knifemaking and bladesmithing data is real-world, first-person video of a person doing the metal craft the catalog row describes: forging, grinding, heat treating and finishing blades in a metal workshop.
• Task familyDevice repair robot training data →Egocentric electronics and device repair video data is real-world, first-person video of a person diagnosing a fault, opening a device, replacing a screen, battery or board and testing the result, captured with 3D hand pose, depth and frame-accurate action segments on a written spec.
• Task familyE-waste disassembly robot training data →Egocentric e-waste disassembly data is real-world, first-person video of a person taking electronic devices apart for recycling, which the site states as "Unscrew and pry open devices, remove boards and batteries and sort materials into bins," captured with 3D hand pose, depth and frame-accurate action segments on a written spec.
• Task familyPlant potting robot training data →Egocentric plant potting and nursery data is real-world, first-person video of a person potting and transplanting on greenhouse and nursery benches, which the site states as two catalog rows, "Pot plants" ("Fill a pot, plant, and water") and "Transplant seedlings" ("Lift and replant seedlings"), captured with 3D hand pose, depth and frame-accurate action segments on a written spec.
• Task familyNursery robot training data →Egocentric seed tray sowing and transplanting data is real-world, first-person video of a person placing seeds into tray cells and lifting and replanting seedlings on greenhouse and nursery benches, which the site states as two catalog rows, "Seed a tray" ("Place seeds into tray cells") and "Transplant seedlings" ("Lift and replant seedlings"), captured with 3D hand pose, depth and frame-accurate action segments on a written spec.
• Task familyPruning robot training data →Egocentric pruning and trellis training data is real-world, first-person video of a person pruning, staking and training plants, which the site states as five catalog rows: "Prune with shears," "Stake + clip plant," "Trellis training" and "Trim + deadhead" under Greenhouse, Nursery & Garden Centre, and "Pruning" under Farm & Field Operations.
• Task familyJanitorial robot training data →Egocentric commercial janitorial data is real-world, first-person video of a person doing janitorial and facilities cleaning in offices and commercial buildings, which the site states as a catalog row, "Commercial Cleaning & Janitorial" ("Janitorial and facilities cleaning across offices and commercial buildings"), alongside 14 neighbouring cleaning rows.
• Task familyBicycle repair robot training data →Egocentric bicycle repair video data is real-world, first-person video of a person adjusting brakes and gears, replacing tubes, chains and cables and truing wheels, captured with 3D hand pose, depth and frame-accurate action segments on a written spec.
• Task familyMailroom robot training data →Egocentric mailroom and document handling video data is real-world, first-person video of a person opening, sorting and distributing mail, filing and scanning documents, collating and stapling sets, and stuffing and sealing envelopes, captured with 3D hand pose, depth and frame-accurate action segments on a written spec.
• Task familyOffice robot training data →Egocentric office printer and shredder data is real-world, first-person video of a person loading a paper tray, clearing a jam and feeding sheets into a shredder, captured with depth, 3D hand pose and frame-accurate action segments on a written spec.
• Task familyEnvelope stuffing robot training data →Egocentric envelope stuffing and mailer packing data is real-world video of a person inserting documents into envelopes and sealing them, and packing a mailer and taping it shut, captured with depth, 3D hand pose and frame-accurate action segments on a written spec.
• Task familyBartending robot training data →Egocentric bartending video data is real-world, first-person video of a person building and garnishing a bar drink, pouring a drink and serving it to a place, and working a barista station, captured with 3D hand pose, depth and frame-accurate action segments on a written spec.
• Task familyShoe repair robot training data →Egocentric shoe repair video data is real-world, first-person video of a person removing worn soles or heels, cutting and gluing replacements, and stitching and finishing edges, captured with 3D hand pose, depth and frame-accurate action segments on a written spec.
• Task familyWindow cleaning robot training data →Egocentric window cleaning video data is real-world, first-person video of a person wetting windows with a pole applicator, pulling the water off with a squeegee, cleaning interior and exterior glass and polishing edges, captured with 3D hand pose, depth and frame-accurate action segments on a written spec.
• Task familyTile installation robot training data →Egocentric tile and flooring installation video data is real-world, first-person video of a person spreading adhesive, setting tiles with spacers, cutting edge tiles and grouting the joints, captured with 3D hand pose, depth and frame-accurate action segments on a written spec.
• Task familyGift wrapping robot training data →Egocentric gift wrapping video data is real-world video of a person wrapping an item in paper and folding it, captured with 3D hand pose, depth and frame-accurate action segments against a written spec.
• Task familyFood assembly robot training data →Egocentric sandwich and salad assembly video data is real-world video of a person assembling food by hand, such as a salad or an order of burgers, wraps and sides, captured with 3D hand pose, depth and frame-accurate action segments against a written spec.
• Task familyEgocentric navigation dataset →Egocentric walking and navigation video data is real-world video of a person moving through a space while wearing the capture rig, such as walking a city street or riding public transit, captured with frame-accurate action segments against a written spec.
• Task familyHome repair robot training data →Egocentric home repair and handyman video data is real-world video of a person doing a repair call in a client’s home while wearing the capture rig, such as a handyman, electrical, HVAC or plumbing call, captured with frame-accurate action segments against a written spec.
• Task familyHVAC technician training data →Egocentric HVAC service video data is real-world video of a technician doing an HVAC service call while wearing the capture rig, such as inspecting a unit, replacing filters or parts, tuning the system or wiring a thermostat, captured with frame-accurate action segments against a written spec.
• Task familyapparel manufacturing robot training data →Egocentric garment production video data is real-world video of a person doing garment or textile production work while wearing the capture rig, such as cutting fabric, sewing seams, pressing finished garments or loading mill machines, captured with frame-accurate action segments against a written spec.
• Task familyarts and crafts robot training data →Egocentric arts and crafts video data is real-world video of a person doing studio art work while wearing the capture rig, such as painting on canvas, painting a mural, printmaking or sculpting, captured with frame-accurate action segments against a written spec.
• Task familyDelivery robot training data →Egocentric last-mile package delivery video data is real-world, first-person video of a courier sorting parcels in a vehicle, carrying them to the door, scanning them and handing them to the recipient, captured with 3D hand pose, depth and frame-accurate action segments on a written spec.
• Task familyMail sorting robot training data →Egocentric postal sorting data is real-world, first-person video of a person sorting letters and parcels by route into pigeonholes and trays at a sorting bench, captured with 3D hand pose, depth and frame-accurate action segments on a written spec.
• Task familyEgocentric food delivery data →Egocentric food delivery pickup and handoff data is real-world, first-person video of a courier collecting a food order at the restaurant, bagging it in an insulated carrier and delivering it, captured with depth, 3D hand pose and frame-accurate action segments on a written spec.
• Task familyPottery robot training data →Egocentric pottery and ceramics video data is real-world, first-person video of studio ceramics work, such as wheel throwing, hand-building, ceramic sculpting, glazing and kiln loading, captured with 3D hand pose, depth and frame-accurate action segments on a written spec.
• Task familyJewelry robot training data →Egocentric gem setting and engraving data is custom first-person video of a person seating a stone in its setting and pushing or beading the prongs over it with fine tools, or cutting letters and patterns into metal with gravers or a rotary tool under magnification, captured by a paid, consented contributor wearing a capture rig and doing the task named in a written skill spec.
• Task familyJewelry making robot training data →Egocentric jewelry and metalsmithing data is real-world, first-person video of bench jewelry work, such as engraving, gem setting, soldering, wire wrapping and polishing, captured with 3D hand pose, depth and frame-accurate action segments on a written spec.
• Task familyLeathercraft robot training data →Egocentric leathercraft data is real-world, first-person video of leatherworking at a bench, such as hand-stitching leather goods, making a bag or wallet, tooling a pattern and working saddlery, captured with 3D hand pose, depth and frame-accurate action segments on a written spec.
• Task familyLandscaping robot training data →Egocentric landscaping and groundskeeping data is real-world, first-person video of grounds work, such as mowing, trimming, planting, clearing paths and pruning hedges, captured with 3D hand pose, depth and frame-accurate action segments on a written spec.
• Task familyData center robot training data →Egocentric data center and IT hardware task data is real-world video of server rack and hardware bench work, such as patching cables, swapping hot-swap modules, seating memory and mounting rails, recorded to a written spec.
• Task familyPackaging line video dataset →Egocentric print and packaging line data is real-world video of work on a packaging line, such as packing cartons, sealing and labeling packs, palletizing and wrapping or folding, recorded to a written spec.
• Task familyFast food robot training data →Egocentric fast food and quick-service kitchen data is real-world video of work in a quick-service kitchen or at its counter, such as taking and handing over orders, running the fryer and the grill, and assembling and wrapping orders, recorded to a written spec.
• Task familyProduce packing robot training data →Egocentric produce packing and grading video data is real-world, first-person video of produce sorting, weighing, packing and grading work, such as cleaning and sorting strawberries, packing them into clamshell containers and grading crops into crates, captured with 3D hand pose, depth and frame-accurate action segments on a written spec.
• Task familyFood manufacturing robot training data →Egocentric food and beverage manufacturing video data is real-world, first-person video of plant work, such as inspecting and pulling defective products, operating and monitoring a filling line, sealing and labelling packs and washing down equipment, captured with 3D hand pose, depth and frame-accurate action segments on a written spec.
• InspectionFood inspection robot training data →Egocentric food QA and foreign object inspection data is real-world, first-person video of a person holding products under an inspection light, inspecting items on the belt for foreign objects and defects and removing anything suspect, captured with 3D hand pose, depth and frame-accurate action segments on a written spec.
• PackagingHeat sealing robot training data →Egocentric heat sealing and labeling data is real-world, first-person video of a person feeding packs into a heat sealer, checking the seal and running the packs through a label applicator, captured with 3D hand pose, depth and frame-accurate action segments on a written spec.
• Filling line operationFilling line robot training data →Egocentric filling line operation data is custom first-person video of people starting a filling line, reading gauges, adjusting settings on the panel and clearing minor jams, and of feeding product onto a line and monitoring the machines, captured by paid, consented contributors wearing a capture rig and doing the task named in a written skill spec.
• Slide prep and plate streakingPlate streaking robot training data →Egocentric slide prep and plate streaking data is custom first-person video of people dropping, smearing and covering a slide, and of swabbing a sample and streaking a plate, captured by paid, consented contributors wearing a capture rig and doing the task named in a written skill spec.
• Task familyCatering robot training data →Egocentric catering and event kitchen video data is real-world, first-person video of kitchen work for events and institutions, such as portioning dishes onto plates in sequence, prepping vegetables, proteins and sauces in large batches, and setting out and clearing a service area, captured with 3D hand pose, depth and frame-accurate action segments on a written spec.
• Task familyGym robot training data →Egocentric gym and fitness facility task video data is real-world, first-person video of the upkeep work in a gym or fitness space, such as setting up benches, racks and machines for a class, mopping floors and wiping down equipment, and returning weights to the racks in order, captured with 3D hand pose, depth and frame-accurate action segments on a written spec.
• Task familyEvent setup robot training data →Egocentric event venue setup and teardown video data is real-world, first-person video of the work that turns an empty room into an event layout and back again, such as arranging and aligning tables and chairs, setting up and tearing down staging and equipment, routing and taping cables, and sweeping up afterwards, captured with 3D hand pose, depth and frame-accurate action segments on a written spec.
• Task familySecurity screening robot training data →Egocentric security checkpoint task video data is real-world video of the hand work done at a staged checkpoint, such as loading screening trays, loading bags to a belt, searching and re-packing a bag, retrieving a flagged item and stacking bins, captured on a mock set with a multiview rig and handheld grippers against a written spec.
• Task familyLuggage handling robot training data →Egocentric luggage and baggage handling data is real-world video of the hand work involved in moving bags: loading guest luggage onto a trolley and delivering it to a room in a hotel, and, on a staged mock checkpoint, loading bags to a belt, scanning a bag tag and opening, inspecting and re-packing a bag.
• Task familyPool maintenance robot training data →Egocentric pool and spa maintenance data is real-world, first-person video of the upkeep work in a hotel pool and spa area, as the catalog row Pool & Spa Area Maintenance states it: skim and test the pool, arrange loungers, wipe surfaces and restock towels.
• Task familyVeterinary robot training data →Egocentric veterinary clinic task video data is real-world video of the hand work done in a veterinary clinic, such as preparing an examination room, laying out instruments on a tray, cleaning a kennel, handling an animal and post-procedure care, captured with a head-mounted phone or a multiview rig against a written spec.
• Task familyKennel cleaning robot training data →Egocentric kennel and animal pen cleaning data is real-world, first-person video of the hand work of cleaning animal housing, such as removing bedding and waste, hosing a kennel and laying fresh bedding, or mucking out a pen with a fork and shovel, captured with a head-mounted phone or a multiview rig against a written spec.
• Task familyEgg collection robot training data →Egocentric egg collection and candling data is real-world, first-person video of the hand work of collecting eggs from the nests, checking them against a light for defects and packing them, captured with a head-mounted phone or a multiview rig against a written spec.
• Task familyAppliance repair robot training data →Egocentric appliance repair data is real-world, first-person video of the hand work of repairing an appliance: opening it, testing its components, replacing the faulty part and reassembling it, captured with a head-mounted phone or a multiview rig against a written spec.
• Task familyFloor care robot training data →Egocentric carpet extraction and floor stripping data is real-world, first-person video of two floor care jobs: extracting and drying carpets with a carpet extractor and air movers, and stripping and waxing hard floors with a floor machine or buffer and a wax applicator.
• Task familyFood plant sanitation robot training data →Egocentric food plant sanitation and washdown data is real-world, first-person video of cleaning food-production and kitchen spaces: rinsing equipment with hoses, applying foam sanitizer, scrubbing and rinsing down a production area, and degreasing and scrubbing restaurant kitchen surfaces.
• Task familyIrrigation robot training data →Egocentric irrigation setup and repair data is real-world, first-person video of people moving and connecting irrigation lines, laying out drip or sprinkler lines, cutting PVC pipe, fitting connectors and replacing broken sections.
• Task familyConstruction cleanup robot training data →Egocentric construction site cleanup data is real-world, first-person video of people clearing debris, sorting waste, preparing a site after works, and sweeping and hauling debris with brooms and wheelbarrows.
• Capture tierTactile glove robot training data →Egocentric tactile glove and hand tracking data is first-person video recorded together with a signal from tactile gloves that track the hands, time-synced to a head-mounted stereo camera by hardware timestamps and an IMU.
• SourcingReal worker egocentric data →Egocentric data from real workers is first-person video of people doing their actual jobs at working sites during normal operations, rather than staged or acted demonstrations.
• DeliveryRobot training data delivery →Egocentric data is delivered through a fixed path: devices upload daily over encrypted TLS transfer into a cloud bucket with sha256 checksums at ingest, the data is validated and packaged in your format, then pushed to your S3, GCS or Azure bucket by cross-account transfer, or sent on encrypted drives for very large drops.
• Task familyGardening robot training data →Egocentric plant care and gardening video data is real-world, first-person video of a person looking after plants: watering, potting, pruning, planting, transplanting and picking by hand, captured with 3D hand pose, depth and frame-accurate action segments on a written spec.
• Retail & shelf opsRetail robot training data →Retail robotics training data is real-world, first-person demonstration data of shelf stocking, facing, restocking and picking tasks captured against a written spec: egocentric video with depth, hand pose, and action segments.
• Manufacturing & workshopAssembly task data →Assembly task data is real-world, first-person demonstration data of manufacturing and workshop tasks — fastening, measuring, press-fit, and wire termination — captured against a written spec: egocentric video with depth, 3D hand pose, contact events, and frame-accurate action segments.
• Cable and wiringEgocentric cable routing and wiring video data →Egocentric cable routing and wiring data is custom first-person video of people routing, clipping, dressing, plugging and wiring cables, captured by paid, consented contributors wearing a capture rig and doing the tasks named in a written skill spec.
• Scanning and labelingEgocentric barcode scanning and labeling video data →Egocentric barcode scanning and labeling data is custom first-person video of people scanning items and parcels, printing labels, applying shipping, product or asset labels and tagging merchandise, captured by paid, consented contributors wearing a capture rig and doing the tasks named in a written skill spec.
• Wiping and cleaningEgocentric wiping and surface cleaning video data →Egocentric wiping and surface cleaning data is custom first-person video of people wiping counters, tables and other surfaces, clearing them beforehand and resetting them afterwards, captured by paid, consented contributors wearing a capture rig and doing the tasks named in a written skill spec.
• Floor cleaning and moppingEgocentric floor cleaning and mopping video data →Egocentric floor-cleaning video data is custom first-person video of people mopping, sweeping and vacuuming floors, captured by paid, consented contributors wearing a capture rig and doing the tasks named in a written skill spec.
• Ironing and garment handlingEgocentric ironing and garment handling video data →Egocentric ironing and garment handling data is custom first-person video of people pressing, hanging, rehanging and sorting garments and linens, captured by paid, consented contributors wearing a capture rig and doing the tasks named in a written skill spec.
• Electronics assemblyEgocentric electronics assembly video data →Egocentric electronics assembly data is custom first-person video of people placing components on boards, seating and press-fitting connectors, fastening housings and routing harnesses, captured by paid, consented contributors wearing a capture rig and doing the tasks named in a written skill spec.
• Task familyScrewdriving robot training data →Egocentric screwdriving and fastening data is real-world, first-person video of a person driving screws, torquing bolts and seating other fasteners with a hand tool or power driver, captured with 3D hand pose, depth, contact events and frame-accurate action segments on a written spec.
• Painting and coatingEgocentric painting and coating video data →Egocentric painting and coating video data is custom first-person video of people applying paint, primer, stain or another finish to a surface with brushes, rollers and rags, captured by paid, consented contributors wearing a capture rig and doing the tasks named in a written skill spec.
• Woodworking and carpentryEgocentric woodworking and carpentry video data →Egocentric woodworking and carpentry video data is custom first-person video of people measuring, sawing, drilling, joining, sanding and finishing wood, captured by paid, consented contributors wearing a capture rig and doing the tasks named in a written skill spec.
• Veneer lamination and pressingVeneer lamination robot training data →Egocentric veneer lamination and pressing data is custom first-person video of people cutting veneer sheets, spreading adhesive, positioning the sheets on a substrate and pressing them flat, captured by paid, consented contributors wearing a capture rig and doing the task named in a written skill spec.
• Furniture assemblyEgocentric furniture assembly video data →Egocentric furniture assembly video data is custom first-person video of people assembling furniture from components, such as a flat-pack panel unit, a glued and screwed frame or a commercial furniture unit, captured by paid, consented contributors wearing a capture rig and following a written skill spec.
• Bathroom cleaningEgocentric bathroom cleaning video data →Egocentric bathroom cleaning video data is custom first-person video of people cleaning bathrooms and restrooms, such as scrubbing a toilet, sink and shower, wiping mirrors and restocking supplies, captured by paid, consented contributors wearing a capture rig and following a written skill spec.
• Meal plating and servingEgocentric meal plating and serving video data →Egocentric meal plating and serving video data is custom first-person video of people plating dishes, portioning food and serving drinks or trays, captured by paid, consented contributors wearing a capture rig and following a written skill spec.
• Food portioning and weighingEgocentric food portioning and weighing data →Egocentric food portioning and weighing data is custom first-person video of people weighing food into portions, filling containers to spec, and sealing, labeling and stacking them for service, captured by paid, consented contributors wearing a capture rig and following a written skill spec.
• Plumbing and pipe fittingEgocentric plumbing and pipe fitting video data →Egocentric plumbing and pipe fitting video data is custom first-person video of people cutting, joining and installing pipe, fitting and replacing fixtures and clearing drains, captured by paid, consented contributors wearing a capture rig and following a written skill spec.
• Trade service callsEgocentric water heater replacement data →Egocentric water heater replacement data is custom first-person video of a person draining and disconnecting an old water heater, positioning the new unit, connecting water and power and testing, captured by a paid, consented contributor wearing a capture rig and following a written skill spec.
• Carpentry and furniture workshopEgocentric cabinet hardware installation data →Egocentric cabinet hardware installation data is custom first-person video of a person marking, drilling and fitting hinges, handles and slides onto cabinets and furniture, using a drill and hinge jig where the task names them, captured by a paid, consented contributor wearing a capture rig and following a written skill spec.
• Trade service callsEgocentric drain snaking and clearing data →Egocentric drain snaking and clearing data is custom first-person video of a person feeding a drain auger into a line, working through the blockage, flushing the drain and checking the line with a light, captured by a paid, consented contributor wearing a capture rig and following a written skill spec.
• Plastic and light assemblyEgocentric injection molding machine tending data →Egocentric injection molding machine tending data is custom first-person video of a person tending an injection-molding machine, opening the mold press, pulling parts with gloves, checking them, trimming and placing them in part bins and restarting the cycle, captured by a paid, consented contributor wearing a capture rig and following a written skill spec.
• Inventory counting and stock checkEgocentric inventory counting and stock check video data →Egocentric inventory counting and stock check video data is custom first-person video of people counting stock in assigned locations, scanning shelf locations and reconciling the totals, captured by paid, consented contributors wearing a capture rig and following a written skill spec.
• Pet careEgocentric pet care video data →Egocentric pet care video data is custom first-person video of people grooming, handling and caring for animals and cleaning their spaces, captured by paid, consented contributors wearing a capture rig and following a written skill spec.
• Pet grooming salonPet grooming robot training data →It is custom video of paid, consented contributors doing grooming-salon work, captured to a written skill spec and delivered in your training format with consent and anonymization records and a commercial training license.
• Hair salon and beautyEgocentric hair salon and beauty task video data →Egocentric hair salon and beauty task video data is custom first-person video of people cutting, coloring, washing, drying and styling hair and doing nail and skin-care work, captured by paid, consented contributors following a written skill spec, with synchronized streams, 3D hand pose, action segments and a commercial training license.
• Task familySalon robot training data →Egocentric hair coloring data is first-person video of a person mixing colour, sectioning the hair, applying it with a brush and foils, timing it and rinsing, recorded from a head-mounted view with two wrist cameras, depth, 3D hand pose and action segments.
• Task familyBeauty robot training data →Egocentric makeup and lash application data is real-world, first-person video of consented participants applying professional makeup, doing lash extension, tinting and shaping, styling brows and giving basic skin and beauty treatments in a salon or studio.
• Rope, cord and laceEgocentric knot tying and rope handling data →Egocentric knot tying and rope handling data is custom first-person video of people handling laces and cables, captured by paid, consented contributors following a written skill spec, with synchronized streams, 3D hand pose, contact events, action segments and a commercial training license.
• Pallet building and wrappingPallet building robot training data →Egocentric pallet building and wrapping data is custom first-person video of people stacking boxes onto pallets, breaking pallets down and stretch-wrapping loads, captured by paid, consented contributors following a written skill spec, with synchronized streams, 3D hand pose, contact events, action segments and a commercial training license.
• Vehicle maintenanceVehicle maintenance robot training data →Egocentric vehicle maintenance data is custom first-person video of people changing oil, replacing filters and batteries, servicing brakes and running diagnostics, captured by paid, consented contributors following a written skill spec, with synchronized streams, 3D hand pose, contact events, action segments and a commercial training license.
• Sewing and textile handlingEgocentric sewing and textile handling video data →Egocentric sewing and textile handling video data is custom first-person video of people cutting, pinning, hemming, stitching and handling fabric, captured by paid, consented contributors wearing a capture rig and doing the tasks named in a written skill spec.
• Bed making and linen changesEgocentric bed making video data →Egocentric bed making video data is custom first-person video of people making beds and changing linens, captured by paid, consented contributors wearing a capture rig and doing the tasks named in a written skill spec.
• Baking and dough handlingEgocentric baking and dough handling video data →Egocentric baking and dough handling video data is custom first-person video of people kneading dough, rolling it flat, measuring ingredients and mixing a batter, captured by paid, consented contributors wearing a capture rig and doing the tasks named in a written skill spec.
• Manufacturing & workshopEgocentric tool use data →Egocentric tool use data is real-world, first-person video of a person handling and operating hand tools — selecting a driver or bit, clamping and aligning a workpiece, measuring to tolerance, fastening, and handing a tool between hands — captured with depth, 3D hand pose, contact events, and frame-accurate action segments on a written spec.
• Manipulation primitivePick and place training data →Egocentric pick and place data is real-world, first-person video of a person reaching for, grasping, transporting, and placing an object, captured with 3D hand pose, contact events, depth, and frame-accurate action segments on a written spec.
• Capture settingTabletop manipulation dataset →Egocentric tabletop manipulation data is first-person video of a person performing manipulation tasks on a single work surface, captured with 3D hand pose, contact events, depth and action segments against a written spec.
• Task familyObject rearrangement dataset →Egocentric object rearrangement and tidying data is first-person video of a person moving many objects from where they are to where they belong: putting items away, sorting and restoring a scene to a goal arrangement.
• Task familyDishwashing robot training data →Egocentric dishwashing video data is first-person video of a person washing, racking, drying and putting away dishes and clearing a kitchen after use.
• Contact labelsHand-object contact data →Egocentric hand-object contact data is first-person video annotated with contact events: timestamped labels for when fingers make and break contact with an object.
• Wrist-view captureWrist camera data collection →Wrist camera data collection is recording human demonstrations through cameras worn at the wrists, so the video moves with each hand rather than staying on the head or in the room.
• Failure and recoveryFailure recovery demonstrations →Failure and recovery demonstration data is first-person human video in which the hard parts of a task, such as a dropped object, an occluded grasp or a re-grasp, are kept in the episode instead of being cut out, and the recovery that follows is labeled as its own action segment.
• Manipulation primitiveGrasping training data egocentric →Egocentric grasping data is first-person video of a human grasping an object, captured with 3D hand pose, contact events, and depth on a written skill spec.
• PricingHow much does custom robotics training data cost? →There is no fixed price list, because cost is driven by what a spec asks for, not a catalog item.
• Video annotationEgocentric video annotation →Egocentric video annotation is labeling first-person (head- or wrist-mounted) video with the structured signals a robot policy trains on — 3D hand pose, action-segment boundaries, contact and grasp events, and object or instance masks — tied to the synchronized depth stream and hardware clock the video was recorded on, rather than a caption or single classification tag per clip.
• Image & video annotationImage and video annotation →Image and video annotation for robotics is labeling camera footage with the spatial signals a perception or policy model trains on — bounding boxes, polygon outlines, per-pixel segmentation masks, object classes and instance identities that persist across frames — tied to the depth and timing of the original capture.
• Keypoint & pose annotationKeypoint and pose annotation →Keypoint and pose annotation is labeling the positions of specific points on a body or hand — joints such as knuckles, wrists and fingertips — in every frame of a video, so a model can learn how a person moves.
• Quality & validationData labeling and quality verification →Data labeling quality verification is the process of checking annotations and footage against a defined acceptance rule before they are delivered — reviewing labels, rejecting items that fail, and recording why — so a buyer receives a dataset that trains rather than one that has to be filtered later.
• Quality & validationValidated egocentric video data →Validated egocentric video data is first-person footage that has passed a written acceptance test — sync within the measured ceiling, calibration solved and within drift tolerance, coverage cells filled against the spec, and consent documented — before it counts toward a delivery.
• Human activityHuman activity video data →Human activity video data is footage of people performing real tasks, used to train models that need to recognize or imitate what a body does.
• Manipulation signalHand-object interaction data →Hand-object interaction video data is footage of a hand approaching, contacting, and manipulating an object, captured with 3D hand pose and contact-event annotation synchronized to a depth stream on one hardware clock, so a manipulation policy can learn the approach, grasp, and release rather than just seeing a hand near an object in a single frame.
• Hand trackingEgocentric hand tracking data →Egocentric hand tracking data is video captured from a first-person viewpoint with the 3D position of hand and finger joints tracked frame by frame, typically delivered as a per-frame joint skeleton with a visibility flag on each joint, so a model can learn exactly how the hand is shaped and moved, including through occlusion, rather than inferring it from a single 2D frame.
• Wrist trajectoriesEgocentric video with wrist trajectories →Egocentric video with wrist trajectories is first-person task footage delivered together with the motion of the wearer’s wrists over time.
• Body pose collectionEgocentric video with body pose data →Egocentric video with body pose data is first-person task footage delivered together with the 3D positions of the wearer’s body joints over time.
• Segmentation mask collectionEgocentric video with segmentation masks →Egocentric video with segmentation masks is first-person task footage delivered with per-pixel masks that say which object each pixel belongs to, with the same object keeping the same ID across the episode.
• Eye gazeEgocentric video with eye gaze data →Egocentric video with eye gaze data is first-person task footage delivered with a record of where the wearer’s attention was directed.
• Audio and videoEgocentric audio and video data collection →Egocentric audio and video data collection is the custom recording of first-person task footage together with an audio stream from the same moment.
• POV videoPOV video data collection →POV video data collection is recording video from the point of view of the person doing a task, using a head, chest, or glasses-mounted camera, so the footage shows hands, tools, and objects the way the actor sees them.
• DocumentationEgocentric dataset with a data card →An egocentric dataset delivered with a data card is first-person video collected to a buyer’s written spec and shipped with a structured record of how it was collected.
• Multi-countryGlobal egocentric data collection →Multi-country egocentric data collection is first-person video recorded by contributors in more than one country under a single written skill spec, so that geography is planned as a diversity axis rather than left to whoever was easiest to recruit.
• IDM and latent actionsEgocentric data for inverse dynamics models →It is custom first-person video, delivered with the action-related fields captured alongside it, that an inverse dynamics model (IDM) or latent action model team can train on, calibrate against or evaluate with, instead of relying only on actions inferred after the fact.
• Sample requestRequest an egocentric data sample →An egocentric data sample is a set of real first-person recordings, with the sensor and label data that ship beside them, supplied so a buyer can evaluate Firsthand data before commissioning a custom collection.
• Spec-driven collectionSkill-spec-driven egocentric data collection →Skill-spec-driven egocentric data collection is custom first-person data capture scoped by a written skill spec, the contract that names the skill, the environments and conditions, the acceptance criteria and the target hours.
• Calibrated collectionCalibrated egocentric video data →Calibrated egocentric video data is first-person task footage delivered together with the calibration of the cameras that recorded it: per-camera intrinsics, extrinsics that place the other cameras relative to the head camera, and per-stream timestamps on one shared clock.
• VLA pretrainingVLA pretraining data →A VLA model’s pretraining stage needs broad, diverse egocentric video paired with action labels and language annotation at scale, so the model learns a general visual and physical prior before fine-tuning specializes it on any one task.
• World-model pretrainingWorld model pretraining data →A robot world model’s pretraining stage needs dense, diverse egocentric video paired with a synchronized action, pose, or depth signal at scale, so the model learns to predict how a scene changes in response to an action before any downstream policy is trained on it.
• Humanoid pretrainingHumanoid pretraining data →A humanoid foundation model’s pretraining stage needs broad, diverse egocentric video captured from a head-worn viewpoint with both hands and whole-body pose in frame, spanning many homes, workplaces, tasks, and people, so the model learns a general bimanual and locomotion prior before fine-tuning specializes it on any one task.
• Humanoid mid-trainingHumanoid mid-training data →Humanoid mid-training data is a smaller, embodiment-specific dataset collected between broad humanoid pretraining and task-specific fine-tuning.
• Mid-trainingmid-training data →Mid-training data is a smaller, embodiment-specific dataset collected between broad pretraining and task-specific fine-tuning.
• Foundation model datarobot foundation model training data →Training data for a robot foundation model spans three stages that narrow in scope as training progresses: broad pretraining data captures general human behavior across many skills and settings; mid-training data pairs matched human and robot capture aligned to one target embodiment; and post-training (fine-tuning) data is scoped to the exact deployment task on the exact deployment robot.
• Post-trainingRobot policy fine-tuning data →Fine-tuning data (also called post-training data) is a small, precisely scoped dataset captured against one deployment task, robot, and environment — including the failure and recovery cases a policy has to handle — used to specialize an already-pretrained (and, in some recipes, mid-trained) model rather than to build general coverage from scratch.
• VLA fine-tuningVLA fine-tuning data →VLA fine-tuning data is a small, task-specific dataset used to adapt an already-pretrained vision-language-action model to your tasks, objects and setting.
• PretrainingBimanual manipulation pretraining data →Bimanual manipulation pretraining data is broad-coverage, egocentric human video of two-handed tasks, captured with per-hand 3D pose, contact events, and depth, recorded across a wide range of objects, environments, and coordination patterns rather than one narrow skill.
• Pretraining dataHousehold robot pretraining data →Household robot pretraining data is large-scale, general-purpose egocentric capture of home tasks — laundry, tidying, cleaning, and surface reset — used to train the base layer of a household robot foundation model before it is aligned to any one home, room, or robot.
• Pretraining dataIndustrial robot pretraining data →Industrial robot pretraining data is large-scale, general-purpose egocentric capture of factory, workshop, and warehouse tasks ��� pick and pack, fastening, measuring, and pallet-build �� used to train the base layer of an industrial or warehouse robot foundation model before it is aligned to any one site, station, or robot.
• Automotive serviceAutomotive robot training data →Automotive service robotics training data is real-world, first-person demonstration data of wheel, fluid, trim and diagnostic tasks in a working service bay, captured against a written spec: egocentric video with depth, hand pose, and action segments.
• Pretraining dataKitchen robot pretraining data →Kitchen task pretraining data is large-scale, general-purpose egocentric capture of food prep, cleaning, and loading and unloading tasks in real kitchens, used to train the base layer of a robot foundation model before it is aligned to any one kitchen, task, or robot.
• Cooking videoEgocentric cooking video data collection →Egocentric cooking video data collection is the custom capture of first-person video of people cooking and preparing food, by paid, consented contributors wearing a capture rig, doing the tasks named in a written skill spec.
• Cutting and food prepEgocentric cutting and food prep video data →Egocentric cutting and food prep data is custom first-person video of people slicing, chopping, dicing and peeling food, captured by paid, consented contributors wearing a capture rig and doing the tasks named in a written skill spec.
• Laundry foldingLaundry folding video data for robot training →Laundry folding video data for robot training is the custom capture of first-person video of people folding and sorting laundry, by paid, consented contributors wearing a capture rig, doing the tasks named in a written skill spec.
• Commercial laundry and linenCommercial laundry and linen handling video data →It is custom first-person video of paid, consented contributors sorting, loading, folding and pressing linen in commercial laundry and hotel linen settings, captured to a written skill spec and delivered in your training format with consent and anonymization records and a commercial training license.
• Dental lab and medical device workshopDental lab and medical device workshop video data →It is custom first-person video of paid, consented contributors doing dental laboratory and medical device workshop tasks, captured to a written skill spec and delivered in your training format with consent and anonymization records and a commercial training license.
• Dental clinicDental clinic task video data →It is custom first-person video of paid, consented contributors doing dental clinic tasks, captured to a written skill spec and delivered in your training format with consent and anonymization records and a commercial training license.
• Sanding and finishingSanding and surface finishing video data →It is custom first-person video of paid, consented contributors sanding, polishing, buffing and deburring, captured to a written skill spec and delivered in your training format with consent and anonymization records and a commercial training license.
• Measuring and markingMeasuring and marking task video data →It is custom first-person video of paid, consented contributors measuring, marking and laying out work, captured to a written skill spec and delivered in your training format with consent and anonymization records and a commercial training license.
• Drilling and mountingDrilling and wall mounting video data →It is custom first-person video of paid, consented contributors drilling holes and mounting or installing hardware, captured to a written skill spec and delivered in your training format with consent and anonymization records and a commercial training license.
• Pharmacy and lab benchPharmacy and lab bench task video data →It is custom first-person video of paid, consented contributors doing pharmacy and laboratory bench tasks, captured to a written skill spec and delivered in your training format with consent and anonymization records and a commercial training license.
• Glass and mosaic craftGlass and mosaic craft video data →It is custom first-person video of paid, consented contributors doing glass and mosaic studio work, captured to a written skill spec and delivered in your training format with consent and anonymization records and a commercial training license.
• Farm and field tasksFarm and field task video data →It is custom first-person video of paid, consented contributors doing farm and field work, captured to a written skill spec and delivered in your training format with consent and anonymization records and a commercial training license.
• Livestock handlingLivestock handling video data →It is custom first-person video of paid, consented contributors doing livestock feeding, watering and pen-cleaning work, captured to a written skill spec and delivered in your training format with consent and anonymization records and a commercial training license.
• Cart and basket retrievalCart retrieval robot training data →It is custom first-person video of paid, consented contributors gathering shopping carts and baskets and returning them to the cart bay or entrance, captured to a written skill spec and delivered in your training format with consent and anonymization records and a commercial training license.
• Returns processingReturns processing robot training data →It is custom first-person video of paid, consented contributors processing returned items in a warehouse or a store, captured to a written skill spec and delivered in your training format with consent and anonymization records and a commercial training license.
• Fitting room and apparel merchandisingApparel retail robot training data →It is custom first-person video of paid, consented contributors resetting fitting rooms, sizing and hanging garments, folding and stacking apparel and facing products in a retail store, captured to a written skill spec and delivered in your training format with consent and anonymization records and a commercial training license.
• Plastic and light assemblyLight assembly manufacturing video data →It is custom first-person video of paid, consented contributors doing light assembly, trimming, inspection and packing work, captured to a written skill spec and delivered in your training format with consent and anonymization records and a commercial training license.
• Part trimming and deflashingDeflashing robot training data →Egocentric part trimming and deflashing data is custom first-person video of people trimming flash from molded parts with flush cutters and cleaning the edges with a deburring blade, captured by paid, consented contributors wearing a capture rig and doing the task named in a written skill spec.
• Task familyFurniture manufacturing robot training data →Egocentric plastic chair manufacturing data is custom first-person video of a person unloading molded chair shells, trimming flash, fitting legs and hardware and stacking finished chairs, captured by a paid, consented contributor wearing a capture rig and doing the task named in a written skill spec.
• Precision assembly benchPrecision assembly robot training data →It is custom video of paid, consented contributors doing small-part bench assembly, captured to a written skill spec and delivered in your training format with consent and anonymization records and a commercial training license.
• Insertion tasksPeg-in-hole robot training data →It is custom first-person video of paid, consented contributors inserting, seating and snapping small parts together at a bench, captured to a written skill spec and delivered with synchronized streams, hand pose, contact events and action segments, consent and anonymization records, and a commercial training license.
• Office and desk tasksEgocentric video data of office and desk tasks →Egocentric video data of office and desk tasks is custom first-person video of people doing everyday office work — desk work with documents, a keyboard and a phone, filing, mail handling, printing, scanning, and desk-and-document station tasks — captured by paid, consented contributors against a written skill spec and delivered with synchronized streams, hand pose, action segments, anonymization and a commercial training license.
• Vision-language modelsEgocentric video for VLM training →Egocentric video data for vision-language model training is first-person video recorded to a written spec and paired with language, so a model can learn how what a person says or is asked to do relates to what the hands and objects in view are doing.
• Pretraining dataBuy robot pretraining data →Yes.
• StartupsPretraining data for robotics startups →Pretraining data for a robotics startup is custom, on-demand egocentric data — first-person human video, depth, hand pose, and action labels — captured to a written spec and delivered under a buyer-owned commercial license, sized to fit an early-stage team’s budget and roadmap rather than requiring the volume or in-house capture program a larger lab would run.
• OrderingOrder a custom egocentric dataset →You order a custom egocentric dataset by writing a spec that defines the tasks, scenes, objects, contributors, coverage, labels, and delivery format you need, getting it quoted, proving it at small scale with a Spec Pilot, then scaling capture.
• Request a quoteGet a quote for egocentric data collection →You get a quote by writing a spec that defines the tasks, scenes, contributors, streams, labels, accept criteria, delivery format, and timeline you need, then submitting it for scoping.
• ScaleLarge-scale egocentric data collection →Large-scale egocentric data collection means executing the same validated first-person capture spec — same streams, same calibration, same labels, same QA bar — across enough contributors, scenes, and sessions that the resulting dataset covers a model’s task and environment distribution broadly, rather than narrowly sampling it.
• MultimodalEgocentric video data for multimodal AI →Egocentric video data for multimodal AI is first-person video, captured from the wearer’s own viewpoint, paired with synchronized streams and labels — such as depth, IMU, and hand or action labels — so a multimodal or video model can learn the relationship between what the camera wearer sees, what their hands are doing, and what task is underway.
• Imitation learningEgocentric data collection for imitation learning →Egocentric data collection for imitation learning is the capture of first-person human demonstrations — video from the demonstrator’s own viewpoint, synchronized with action-segment and hand-pose labels — to a written spec, so an imitation-learning or behavioral-cloning policy can learn to reproduce the demonstrated behavior directly from those recordings.
• Pilot collectionEgocentric data collection pilot →An egocentric data collection pilot — a Spec Pilot — is a small, fast first batch of first-person episodes captured against a buyer’s written skill spec, QA’d and delivered with the same reject log and coverage reporting a full collection gets, so the spec and accept criteria can be proven on real footage before scaling to full volume.
• Small batchSmall custom egocentric dataset →A small custom egocentric dataset is a focused batch of first-person episodes captured to a buyer’s written spec — a first batch, a narrow skill, or a pilot-sized order — rather than a standing, large-volume program.
• Multimodal captureEgocentric video with depth and IMU data →Egocentric video with depth and IMU data is first-person video paired with a metric depth stream recovered from a calibrated stereo rig and 6-DoF inertial motion data from head- and wrist-mounted IMUs, all hardware-triggered onto the same clock so every stream is aligned to the same instant rather than approximately synchronized after the fact.
• Workplace captureEgocentric workplace task data →Egocentric video of real workplace tasks is first-person footage of people doing actual job tasks — picking and packing, stocking shelves, servicing vehicles, assembling parts, prepping food — captured on head- and wrist-mounted cameras in the real setting where the work happens, against a written skill spec, with consent documented and bystanders de-identified.
• StartupsEgocentric video data for robotics startups →Egocentric video data for robotics startups is custom first-person capture scoped to an early-stage team’s specific task list and training stack — ordered from a written skill spec, proven on a small Spec Pilot batch, QA’d, and delivered under a perpetual, buyer-owned license in the format the team’s pipeline already consumes, rather than assembled from whichever public research corpus happens to overlap with the roadmap.
• EnterpriseEnterprise egocentric data collection →Enterprise egocentric data collection is a custom first-person capture program scoped to a large organization’s own written skill spec, run through a Spec Pilot before scaling, reviewed against accept criteria in per-batch QA, backed by a documented consent chain and anonymization record, and delivered under a perpetual, buyer-owned commercial license and in a training-ready format — with the provenance and data-processing documentation an enterprise buyer’s procurement and legal teams review before signing.
• Ongoing programOngoing data collection program →An ongoing data collection program is a custom egocentric capture engagement that continues past the first Spec Pilot as a series of batches rather than a single delivery — each batch run against the same written skill spec, updated as new tasks or failure cases are added, checked against the same QA bar, and delivered under the same buyer-owned license as the pilot.
• By industryEgocentric data by industry →Egocentric data collection by industry is the same first-person capture, labeling, QA, and licensed-delivery process applied across different work settings — warehouse and logistics, retail, automotive service, tool use and assembly, kitchen and food prep, and household chores — with the tasks, scenes, and constraints of each setting written into its own skill spec.
• Warehouse & logisticsLogistics and fulfillment egocentric data →Egocentric data for logistics and fulfillment is real-world, first-person demonstration data of the fulfilment flow — receiving, putaway, picking, packing, sortation, and loading ��� captured against a written spec: video with depth, hand pose, and action segments, from the viewpoint a robot or vision model will actually operate from.
• Workflow captureEnterprise workflow video data collection →Enterprise workflow video data is first-person video of a real, hands-on business process or standard operating procedure, captured against a written skill spec that defines the task’s steps, scenes, and streams, broken into step units through task decomposition, labeled with action segments and hand pose, and delivered under a buyer-owned commercial license in a training-ready format such as RLDS or LeRobot.
• Manufacturing & workshopManufacturing workflow data →Manufacturing workflow data is first-person video of a real, multi-step production or assembly procedure — captured against a written skill spec that decomposes the procedure into step units through task decomposition, labeled with action segments and hand pose across the full sequence, and delivered under a buyer-owned commercial license in a training-ready format such as RLDS or LeRobot.
• Food serviceFood service egocentric data →Egocentric food service data is first-person video and synchronized sensor data of real commercial food-handling tasks — prep and cutting, assembly and plating, cooking-line handling, dishwashing, and packaging — captured in a working kitchen or restaurant setting and labeled into the same skill spec, action segment, episode, and hand pose structure used across every custom collection on this site.
• Construction & tradesConstruction and skilled trades egocentric data →Egocentric construction and trades data is first-person video and synchronized sensor data of real jobsite and skilled-trades tasks, captured on site and labeled into the same skill spec, action segment, episode, and hand pose structure used across every custom collection on this site.
• Procedural tasksProcedural task video data →Procedural task video data is first-person video and synchronized sensor data of a multi-step task performed in a fixed or partly fixed order, captured to a written skill spec and labeled into action segments that mark each step’s boundaries, episodes that group a full procedure attempt, and hand pose, so a model can learn not just that a task was done but how its steps are ordered and where each one starts and ends.
• Long-horizon tasksLong-horizon egocentric task data →Long-horizon egocentric task data is first-person video and synchronized sensor data of a person completing a long, multi-step task end to end, captured to a written skill spec.
• Deformable objectsEgocentric deformable object manipulation data →Egocentric deformable object manipulation data is first-person video and synchronized sensor data of a person handling soft, shape-changing items, such as laundry, bedding and bags, captured to a written skill spec.
• Task familyGarment fastening robot training data →Egocentric buttons, zippers and fasteners data is real-world, first-person video of small fastening tasks on a cloth-and-cable bench, as four catalog rows state them: "Fasten buttons/snaps", "Open/close zipper", "Thread lace eyelets" and "Coil + Velcro cable".
• Task familyHeavy equipment operator training data →Egocentric heavy machinery operation data is real-world, first-person video of a person operating the levers and control panels of presses, cranes or machinery from the operator station, as one catalog row, "Operating Heavy Machinery Controls", states it.
• Task familyStereo machinery operator dataset →Stereo heavy machinery operation data is first-person video from two calibrated cameras of a person working the levers and control panels of presses, cranes or machinery from the operator station, with depth and hand pose alongside.
• Task familySurface finishing robot training data →Stereo surface finishing data is first-person video from two calibrated cameras of a person sanding, taping and mudding drywall, painting or staining, with depth and hand pose alongside, so the surface being finished is recorded in three dimensions.
• Task familyPainting robot training data →Stereo painting and coating data is first-person video from two calibrated cameras of a person priming, rolling, cutting in, brushing or staining a surface, with depth and hand pose alongside, so the wall, edge or panel being coated is recorded in three dimensions.
• Task familyTile installation robot training data →Stereo tile and flooring installation data is first-person video from two calibrated cameras of a person spreading adhesive, setting tiles with spacers, cutting edge tiles and grouting the joints, with depth and hand pose alongside, so the surface being tiled is recorded in three dimensions.
• Task familyDrywall robot training data →Stereo drywall installation and finishing data is first-person video from two calibrated cameras of a person lifting and screwing drywall panels to studs and taping and mudding the seams, with depth and hand pose alongside, so the panel and the joint are recorded in three dimensions.
• Task familyWelding robot training data →Stereo welding and metal fabrication data is first-person video from two calibrated cameras of a person laying out, cutting, fitting, welding and grinding metal parts, with depth and hand pose alongside, so the part, the seam and the fixture are recorded in three dimensions.
• Task familyPlumbing robot training data →Stereo plumbing and pipe fitting data is first-person video from two calibrated cameras of a person cutting and fitting pipe and removing and installing fixtures, with depth and hand pose alongside, so the pipe, the fitting and the fixture are recorded in three dimensions.
• Task familyElectrical installation robot training data →Stereo electrical installation data is first-person video from two calibrated cameras of a person wiring outlets, switches and fixtures, with depth and hand pose alongside, so the box, the device and the conductor are recorded in three dimensions.
• Task familyCarpentry robot training data →Stereo carpentry and framing data is first-person video from two calibrated cameras of a person measuring, marking, cutting and fitting lumber and panels, with hand pose and depth alongside, so the board, the cut line and the work surface are recorded in three dimensions.
• Task familyInsulation installation robot training data →Stereo insulation installation data is first-person video from two calibrated cameras of a person measuring, cutting and fitting insulation batts between studs and joists and stapling them in place, with hand pose and depth alongside, so the batt, the cut line and the framing are recorded in three dimensions.
• Task familyHVAC robot training data →Stereo HVAC installation and service data is first-person video from two calibrated cameras of a technician mounting HVAC units, swapping and tuning filters, wiring a thermostat or working a service call, with hand pose and depth alongside, so the unit, the part and the hands are recorded in three dimensions.
• Task familyGlass manufacturing robot training data →Egocentric glass manufacturing data is real-world, first-person video of the one catalog row on this topic: handling molten or sheet glass and cutting, forming and finishing pieces with tongs, cutters and grinders.
• Task familyRetail backroom robot training data →Egocentric retail backroom organization data is real-world, first-person video of the four catalog rows on this topic: sorting stock onto labelled racks and bins, grouping and shelving backroom stock by category, and inspecting and sorting damaged goods.
• Task familyDamaged goods sorting robot training data →Egocentric damaged goods sorting data is real-world, first-person video of a person inspecting returned or damaged items, scanning them and sorting them into markdown, disposal or returns bins, captured with depth, 3D hand pose and frame-accurate action segments on a written spec.
• Task familyConcrete finishing robot training data →Stereo concrete finishing data is first-person video from two calibrated cameras of a person mixing, pouring, screeding and finishing concrete into forms with shovels, floats and trowels, with depth and hand pose alongside, so the surface being finished is recorded in three dimensions.
• Task familyLayout marking robot training data →Stereo layout marking and taping data is first-person video from two calibrated cameras of a person measuring, marking and laying out work, taping drywall joints or masking an edge, with depth and hand pose alongside, so the surface being marked or taped is recorded in three dimensions.
• Task familyMopping robot training data →Stereo floor mopping data is first-person video from two calibrated cameras of a person mopping and caring for floors, with depth and hand pose alongside, so the floor, the tool and the hands on the handle are recorded in three dimensions.
• Task familyFloor care robot training data →Stereo floor stripping and waxing data is first-person video from two calibrated cameras of a person stripping old wax from a hard floor and laying new wax, with depth and hand pose alongside, so the floor, the machine and the hands on it are recorded in three dimensions.
• Task familyWindow cleaning robot training data →Stereo window and glass cleaning data is first-person video from two calibrated cameras of a person wetting windows with a pole applicator, pulling the water off with a squeegee and cleaning interior and exterior glass, with depth and hand pose alongside, so the pane, the tool and the hands on it are recorded in three dimensions.
• Task familyCar detailing robot training data →Stereo car detailing and polishing data is first-person video from two calibrated cameras of a person applying, buffing and inspecting polishing compound on body panels with a dual-action polisher and microfiber cloths, and washing, claying and drying panels, with depth and hand pose alongside, so the panel, the tool and the hands on it are recorded in three dimensions.
• Task familyCar detailing robot training data →Stereo car interior vacuuming and detailing data is first-person video from two calibrated cameras of a person vacuuming and cleaning vehicle interiors, seats, mats and trim, and vacuuming, wiping and conditioning interiors with a shop vacuum and detailing brushes, with depth and hand pose alongside, so the surface, the tool and the hands on it are recorded in three dimensions.
• Task familyCar detailing wheel dataset →Egocentric wheel and tire dressing data is real-world, first-person video of a person cleaning wheels and tires and dressing them with finishing products, captured with 3D hand pose, depth and frame-accurate action segments on a written spec.
• Task familyEvent staging robot training data →Stereo event staging setup and teardown data is first-person video from two calibrated cameras of a person carrying staging platforms in on hand trucks, locking them together, assembling stage decks, setting up equipment and lighting, then dismantling, stacking and packing them, with depth and hand pose alongside.
• Task familyTire service robot training data →Stereo tire and wheel service data is first-person video from two calibrated cameras of a person dismounting and mounting tires with a tire changer and bead lubricant, balancing wheels on a wheel balancer, and rotating wheels with a pressure check, with depth and hand pose alongside, so the wheel, the machine and the hands on them are recorded in three dimensions.
• Task familyLaundry robot training data →Stereo commercial laundry and linen data is first-person video from two calibrated cameras of a person sorting laundry into carts and bins, loading industrial washers and dryers, folding linens on a table or in a folding machine, and pressing them on a steam press or an industrial press, with depth and hand pose alongside, so the linen, the machine and the hands on them are recorded in three dimensions.
• Task familySewing robot training data →Stereo garment sewing and fabric cutting data is first-person video from two calibrated cameras of a person feeding fabric pieces through an industrial sewing machine and laying out, patterning and cutting fabric with shears or a cutting machine, with depth and hand pose alongside, so the cloth, the machine and the hands on them are recorded in three dimensions.
• Task familyLine cook robot training data →Stereo commercial kitchen line cooking data is first-person video from two calibrated cameras of a person breaking down produce and proteins with knife cuts, chopping and portioning vegetables, and cooking dishes on a professional line, with depth and hand pose alongside, so the blade, the ingredient and the hands on them are recorded in three dimensions.
• Task familyDishwashing robot training data →Stereo dishwashing and kitchen sanitation data is first-person video from two calibrated cameras of a person operating a dish pit, racking and washing commercial kitchenware, and degreasing and scrubbing kitchen surfaces, with depth and hand pose alongside, so the ware, the tool, the surface and the hands on them are recorded in three dimensions.
• Task familyKitchen line setup robot training data →Egocentric kitchen line setup and breakdown data is first-person video of a person filling gastronorm and prep pans with prepped items, arranging and setting them in the line before service, then clearing them down, emptying, cleaning and storing them afterwards, with synchronized streams, hand pose and action segments alongside.
• Task familyShelf stocking robot training data →Stereo retail shelf restocking data is first-person video from two calibrated cameras of a person opening cartons, facing products and filling gaps on shelves, with depth and hand pose alongside, so the carton, the product, the shelf edge and the hands on them are recorded in three dimensions.
• Task familyGlass handling robot training data →Stereo glass cutting and forming data is first-person video from two calibrated cameras of a person handling glass and cutting, forming and finishing pieces with hand tools, with depth and hand pose alongside, so the piece, the tool and the hands on them are recorded in three dimensions.
• Task familyMachine tending robot training data →Stereo injection molding machine tending data is first-person video from two calibrated cameras of a person tending an injection-molding machine, opening the mold press, pulling parts with gloves, checking them, trimming and placing them in part bins and restarting the cycle, with depth and hand pose alongside, so the press, the part and the hands on it are recorded in three dimensions.
• Task familyPacking robot training data →Stereo order packing and taping data is first-person video from two calibrated cameras of a person picking customer orders with a handheld scanner, packing them into boxes, taping the boxes shut with a packing tape dispenser or a tape gun and applying the shipping label, with depth and hand pose alongside, so the box, the tape and the hands on them are recorded in three dimensions.
• Task familyPalletizing robot training data →Stereo palletizing and stretch wrapping data is first-person video from two calibrated cameras of a person boxing finished goods, taping cartons shut with tape guns, stacking them on a pallet in layers and stretch-wrapping the load, with depth and hand pose alongside, so the cartons, the stack and the hands on them are recorded in three dimensions.
• Task familyStereo landscaping robot training data →Stereo landscaping and groundskeeping data is first-person video from two calibrated cameras of a person mowing, trimming, planting and doing general grounds maintenance on commercial property, with depth and hand pose alongside, so the tool, the plant or ground being worked and the hands on the tool are recorded in three dimensions.
• Task familyHousekeeping robot training data →Stereo hotel room turnover and housekeeping data is first-person video from two calibrated cameras of a person resetting a guest room after checkout: stripping and making beds, cleaning bathrooms, dusting, vacuuming, restocking, checking items against the standard and reporting faults, with depth and hand pose alongside.
• Task familyOrder picking robot training data →Stereo warehouse order picking data is first-person video from two calibrated cameras of a person walking a pick path with a handheld scanner and picking items to a tote, and on one row packing and taping the order, with depth and hand pose alongside.
• Task familyE-waste recycling robot training data →Stereo e-waste disassembly and sorting data is first-person video from two calibrated cameras of a person unscrewing and prying open devices, removing boards and batteries and sorting materials into bins, with depth and hand pose alongside.
• Task familyBarista robot training data →Egocentric barista and espresso data is first-person video of a person working a coffee bar: grinding beans, dosing and tamping, pulling espresso shots, steaming milk and pouring drinks, recorded from their own point of view with hand pose, depth and action segments alongside.
• Task familyInventory robot training data →Stereo inventory cycle counting data is first-person video from two calibrated cameras of a person scanning shelf locations, counting stock with a scanner and reconciling the totals, with depth and hand pose alongside.
• Task familyBrake service robot training data →Stereo brake pad and rotor service data is first-person video from two calibrated cameras of a person jacking a car, removing the wheel and caliper, fitting new pads and a rotor and torquing the bolts, with depth and hand pose alongside.
• Task familyOil change robot training data →Stereo oil and fluid change data is first-person video from two calibrated cameras of a person draining engine oil, replacing the oil filter, refilling and topping up fluids, with depth and hand pose alongside.
• Task familySanitation robot training data →Stereo food plant sanitation washdown data is first-person video from two calibrated cameras of a person rinsing production equipment with hoses, applying foam sanitizer, scrubbing and rinsing down a production area, with depth and hand pose alongside.
• Task familyGrinding robot training data →Stereo grinding and deburring data is first-person video from two calibrated cameras of a person grinding weld seams flush with an angle grinder and finishing edges and removing burrs with files, with depth and hand pose alongside.
• Task familyRestroom cleaning robot training data →Stereo restroom cleaning and disinfection data is first-person video from two calibrated cameras of a person cleaning and disinfecting a restroom with disinfectant sprayers and mops, scrubbing toilets, urinals and sinks and restocking paper and soap, with depth and hand pose alongside.
• Task familyCarpet cleaning robot training data →Stereo carpet extraction data is first-person video from two calibrated cameras of a person pre-spraying carpets, running a carpet extractor over them in passes and positioning air movers to dry the floor, with depth and hand pose alongside.
• Task familyWarehouse receiving robot training data →Stereo warehouse receiving and truck loading data is first-person video from two calibrated cameras of a person receiving and unloading inbound pallets with a manual pallet jack, unloading inbound goods from a truck, moving freight to the dock with a hand truck and loading outbound trucks, with depth and hand pose alongside.
• Task familyParcel sorting robot training data →Stereo warehouse sorting and labeling data is first-person video from two calibrated cameras of a person printing labels, applying them to outbound boxes, sorting boxes by carrier or route and sorting and labeling boxes and products by type, with depth and hand pose alongside.
• Task familyProduce packing robot training data →Stereo produce sorting and packing data is first-person video from two calibrated cameras of a person sorting, grading, weighing, packing and labeling fresh produce, such as strawberries into clamshell containers, with depth and hand pose alongside.
• Task familyDevice repair robot training data →Stereo electronics and device repair data is first-person video from two calibrated cameras of a person diagnosing and repairing phones, laptops, tablets and consumer electronics, with depth and hand pose alongside, so the device, the tool and the hands on it are recorded in three dimensions.
• Task familyAssembly robot training data →Stereo consumer goods assembly data is first-person video from two calibrated cameras of a person assembling consumer products on a bench, fitting plastic parts in a jig and fastening them with an electric screwdriver, with depth and hand pose alongside.
• Task familyGarment finishing robot training data →Stereo garment finishing and pressing data is first-person video from two calibrated cameras of a person pressing seams and finished garments on an industrial iron or press and folding or hanging them, with depth and hand pose alongside.
• Task familyFilling line robot training data →Stereo filling line operation data is first-person video from two calibrated cameras of a person starting a filling line, reading gauges, adjusting settings on the panel and clearing minor jams, and of feeding product onto a line and monitoring the machines, with depth and hand pose alongside.
• Task familyFood inspection robot training data →Stereo food QA and foreign-object inspection data is first-person video from two calibrated cameras of a person holding products under an inspection light, inspecting items on the belt for foreign objects and defects and removing anything suspect, with depth and hand pose alongside.
• Task familyFurniture assembly robot training data →Stereo furniture assembly data is first-person video from two calibrated cameras of a person gluing and clamping frame joints, drilling pilot holes and screwing a furniture frame together, or assembling commercial furniture units from components with hand and power tools, with depth and hand pose alongside.
• Task familyCabinet installation robot training data →Stereo cabinet hardware installation data is first-person video from two calibrated cameras of a person marking, drilling and fitting hinges, handles and slides onto cabinets and furniture, using a drill and hinge jig where the row names them, with depth and hand pose alongside.
• Task familyWire harness robot training data →Stereo wire crimping and harness assembly data is first-person video from two calibrated cameras of a person crimping a terminal onto a wire and routing conductors to a form board and clipping them, with depth and hand pose alongside.
• Task familyTorque wrench robot training data →Stereo torque and bolt fastening data is first-person video from two calibrated cameras of a person torquing bolts with a torque wrench and fastening a housing with a driver to torque, with depth and hand pose alongside.
• Task familyCar battery replacement robot training data →Egocentric car battery replacement data is first-person video of a person testing a car battery with a multimeter, disconnecting and swapping it, and cleaning or tightening the terminals, recorded from their own point of view with hand pose, depth and action segments alongside.
• Task familyVehicle diagnostics robot training data →Egocentric vehicle diagnostics data is first-person video of a person connecting a scan tool, reading fault codes, inspecting components and identifying the fault, recorded from their own point of view with hand pose, depth and action segments alongside.
• Task familyCabin filter replacement robot training data →Egocentric air and cabin filter replacement data is first-person video of a person inspecting, removing and installing air and cabin filters with a screwdriver and a shop rag, recorded from their own point of view with hand pose, depth and action segments alongside.
• Task familyVacuuming robot training data →Egocentric vacuuming data is first-person video of a person vacuuming carpets, hard floors and vehicle interiors, recorded from their own point of view with hand pose, depth and action segments alongside.
• Task familyWoodworking robot training data →Stereo table saw and wood cutting data is first-person video from two calibrated cameras of a person measuring and marking wood panels, cutting them on a table saw, shaping curves with a jigsaw, and dimensioning and cutting timber, with depth and hand pose alongside, so the panel, the cut line and the hands on the work are recorded in three dimensions.
• Task familySoldering robot training data →Stereo soldering and circuit board assembly data is first-person video from two calibrated cameras of a person placing components on a board, seating connectors, soldering components with an iron and solder wire, trimming leads and inspecting joints, with depth and hand pose alongside, so the board, the tool and the hands on it are recorded in three dimensions.
• Task familyConstruction site cleanup robot training data →Stereo construction site cleanup data is first-person video from two calibrated cameras of a person clearing debris, sweeping it up and hauling it in a wheelbarrow, with depth and hand pose alongside, so the debris, the tool and the hands on the handle are recorded in three dimensions.
• Task familyCaulking robot training data →Stereo caulking and sealing data is first-person video from two calibrated cameras of a person running a caulk bead and tooling it smooth, with depth and hand pose alongside, so the joint, the bead, the tool and the hands are recorded in three dimensions.
• Task familyHome repair robot training data →Stereo handyman and home repair data is first-person video from two calibrated cameras of a person doing a repair call in a client’s home, with depth and hand pose alongside, so the fixture, the tool and the hands are recorded in three dimensions.
• Task familyData center robot training data →Stereo data center rack and server hardware data is first-person video from two calibrated cameras of a person doing rack and hardware bench work, such as hot-swapping modules, seating memory and mounting rails, with depth and hand pose alongside, so the part, the slot and the hands are recorded in three dimensions.
• Task familySurface quality inspection dataset →Stereo surface quality inspection data is first-person video from two calibrated cameras of a person inspecting parts and items against a standard, with depth and hand pose alongside, so the part, the hands turning it and the station are recorded in three dimensions.
• Recurring supplyEgocentric data supply for robotics labs →A consistent egocentric data supply is a recurring flow of first-person training data in which every batch arrives under the same written spec, the same QA bar, the same delivery format and the same licence.
• Mobile manipulationEgocentric mobile manipulation data collection →Egocentric mobile manipulation data is first-person video and synchronized sensor data of a person who moves between locations and works with objects in the same attempt, captured to a written skill spec.
• Clinical & careHealthcare egocentric data →Egocentric healthcare and hospital data is first-person video and synchronized sensor data of non-clinical hospital tasks, captured only in training environments with simulated patients, written into a skill spec and labeled into action segments, episodes, and hand pose, so a model can learn hospital-adjacent task structure without any real patient footage ever being captured.
• Task familyTable setting robot training data →Egocentric table setting and clearing data is real-world, first-person video of a person laying a place setting, stacking and bussing dishes, and wiping a table, captured with 3D hand pose, depth, contact events and frame-accurate action segments on a written spec.
• Hospitality & housekeepingHospitality egocentric data →Egocentric hospitality and housekeeping data is first-person video of hospitality tasks — guest-facing interactions and back-of-house cleaning or room-turnover work — captured from the worker’s own point of view, written into a skill spec, and labeled into action segments, episodes, and hand pose, so a model can learn hospitality task structure from the same viewpoint a worker uses.
• Hotel room turnoverEgocentric hotel room turnover data →Egocentric hotel room turnover data is first-person video of one person resetting a guest room after checkout, captured by a paid, consented contributor against a written skill spec and delivered as validated episodes with synchronized streams, hand pose and labeled action segments.
• Wood carvingEgocentric wood carving video data →Egocentric wood carving data is first-person video of one person carving or whittling wood, captured by a paid, consented contributor against a written skill spec and delivered as validated episodes with synchronized streams, hand pose and labeled action segments.
• Marquetry and inlayEgocentric marquetry data →Egocentric marquetry and inlay data is custom first-person video of people cutting veneer pieces to pattern and gluing them into a decorative inlay on a panel, captured by paid, consented contributors wearing a capture rig and doing the task named in a written skill spec.
• Woodturning and latheEgocentric woodturning and lathe data →Egocentric woodturning and lathe data is first-person video of one person mounting a blank on a lathe and shaping it with turning tools, captured by a paid, consented contributor against a written skill spec and delivered as validated episodes with synchronized streams, hand pose and labeled action segments.
• Embroidery and hand stitchingEgocentric embroidery and hand stitching data →Egocentric embroidery and hand stitching data is first-person video of one person stitching fabric by hand, captured by a paid, consented contributor against a written skill spec and delivered as validated episodes with synchronized streams, hand pose and labeled action segments.
• Lab pipetting and sample handlingEgocentric lab pipetting and sample handling data →Egocentric lab pipetting and sample handling data is first-person video of one person working at a lab bench, captured by a paid, consented contributor against a written skill spec and delivered as validated episodes with synchronized streams, hand pose and labeled action segments.
• Classroom setup and cleanupEgocentric classroom setup and cleanup data →Egocentric classroom setup and cleanup data is first-person video of one person arranging a room for the day, captured by a paid, consented contributor against a written skill spec and delivered as validated episodes with synchronized streams, hand pose and labeled action segments.
• Childcare and early educationEgocentric childcare and early education task data →Egocentric childcare and early education task data is first-person video of a paid, consented contributor doing care-room work, captured against a written skill spec and delivered as validated episodes with synchronized streams, hand pose and labeled action segments.
• Price tagging and shelf signageEgocentric price tagging and shelf signage data →Egocentric price tagging and shelf signage data is first-person video of a paid, consented contributor attaching price tags, tagging merchandise with a pricing gun and placing shelf signage, captured against a written skill spec and delivered as validated episodes with synchronized streams, hand pose and labeled action segments.
• Electrical workEgocentric electrical work video data →Egocentric electrical work video data is custom first-person video of people doing electrical wiring tasks such as wiring outlets, switches and fixtures, captured by paid, consented contributors wearing a capture rig and working to a written skill spec.
• Drywall and wall finishingEgocentric drywall and wall finishing video data →Egocentric drywall and wall finishing video data is custom first-person video of people doing the drywall tasks the site lists, such as lifting and screwing panels to studs, taping joints and applying joint compound, captured by paid, consented contributors wearing a capture rig and working to a written skill spec.
• Insulation installationEgocentric insulation installation data →Egocentric insulation installation data is custom first-person video of people doing the insulation tasks the site lists, such as measuring and cutting batts with a utility knife, fitting them between studs and stapling them in place, captured by paid, consented contributors wearing a capture rig and working to a written skill spec.
• Masonry and plasteringEgocentric masonry and plastering task data →Egocentric masonry and plastering task data is custom first-person video of people doing hand-tool trade work, captured by paid, consented contributors wearing a capture rig and working to a written skill spec.
• Knitting and crochetEgocentric knitting and crochet video data →Egocentric knitting and crochet video data is custom first-person video of people doing the knitting and crochet tasks the site lists, such as hand and machine knitting of garments and textile products and producing crocheted garments and textile products to commercial order, captured by paid, consented contributors wearing a capture rig and working to a written skill spec.
• Soldering and PCB reworkEgocentric soldering and PCB rework video data →Egocentric soldering and PCB rework video data is custom first-person video of people doing the electronics soldering tasks the site lists, such as soldering components to a board and soldering through-hole components, captured by paid, consented contributors wearing a capture rig and working to a written skill spec.
• Cold food prepEgocentric sushi and cold food prep data →Egocentric sushi and cold food prep data is custom first-person video of people preparing food by hand, such as assembling a salad or plating a portion and adding garnish, captured by paid, consented contributors wearing a capture rig and working to a written skill spec.
• Desk and keyboard workEgocentric computer and keyboard use data →Egocentric computer and keyboard use data is custom first-person video of a person working at a desk, such as the catalog’s desk work row, "Work at a desk handling documents, keyboard and phone through routine office tasks." Paid, consented contributors wear a capture rig and work to a written skill spec.
• Task instruction per episodeInstruction-labeled egocentric task data →Instruction-labeled egocentric task data is custom first-person video in which each episode is paired with one task instruction, recorded by paid, consented contributors working to a written skill spec.
• Instrument handlingSurgical instrument handling robot training data →It is custom first-person video of paid, consented contributors setting up trays, opening sterile packs, arranging instruments and passing them in sequence, captured to a written skill spec and delivered in your training format with consent and anonymization records and a commercial training license.
• Medication handlingMedication handling robot training data →It is custom first-person video of paid, consented contributors sorting blister packs, handling medication from a chart, and dispensing and labeling it, captured to a written skill spec and delivered in your training format with consent and anonymization records and a commercial training license.
• Adhesive and sealant applicationAdhesive dispensing robot training data →It is custom first-person video of paid, consented contributors dispensing an adhesive bead, applying thermal paste, tooling a caulk bead and gluing and clamping a joint, captured to a written skill spec and delivered in your training format with consent and anonymization records and a commercial training license.
• Thermostat and control wiringEgocentric thermostat and control wiring data →Egocentric thermostat and control wiring data is custom first-person video of a person mounting a thermostat, connecting control wires to terminals and testing the heating and cooling calls, captured by paid, consented contributors wearing a capture rig and working to a written skill spec.
• Task familyRetail display robot training data →Egocentric promo display assembly data is first-person video of a person building a small promotional display, captured with 3D hand pose, depth and frame-accurate action segments on a written spec.
• Task familyFast food robot training data →Stereo fast food fryer and grill data is first-person video from two calibrated cameras of a person dropping baskets into a fryer, flipping items on a grill, draining product and assembling orders, with depth and hand pose alongside.
• Task familyQuick service restaurant robot training data →Stereo quick-service order assembly and counter data is first-person video from two calibrated cameras of a person assembling burgers, wraps and sides onto trays and then bagging or boxing items and handing them over at the counter, with depth and hand pose alongside.
• Task familyDelivery robot training data →Stereo parcel and courier delivery data is first-person video from two calibrated cameras of a person sorting parcels in a van, carrying them to the door, scanning them and handing them to the recipient, plus collecting and delivering orders, with depth and hand pose alongside.
• Task familyPipe fitting robot training data →Stereo pipe cutting and fitting data is first-person video from two calibrated cameras of a person measuring and marking steel pipe, cutting it on a bandsaw, and fitting and clamping the sections ready for welding, with depth and hand pose alongside.
• Task familyWeld inspection robot training data →Stereo weld inspection and cleanup data is first-person video from two calibrated cameras of a person inspecting welds for defects, grinding and cleaning seams and marking parts for rework, with depth and hand pose alongside.
• Task familyWater heater installation robot training data →Stereo water heater replacement data is first-person video from two calibrated cameras of a person draining and disconnecting an old water heater, positioning the new unit, connecting water and power and testing, with depth and hand pose alongside.
• Task familyDrain cleaning robot training data →Stereo drain snaking and clearing data is first-person video from two calibrated cameras of a person feeding a drain auger into a line, working through the blockage and flushing the drain or checking the line with a light, with depth and hand pose alongside.
• Task familyAV cable management robot training data →Egocentric AV cable management and taping data is first-person video of a person running cables to AV gear, laying them along walls and walkways, covering them with cable ramps and taping them down with gaffer tape, with hand pose and depth alongside.
• Task familyDeflashing robot training data →Stereo part trimming and deflashing data is first-person video from two calibrated cameras of a person trimming flash from molded parts with flush cutters and cleaning the edges with a deburring blade, with depth and hand pose alongside.
• Task familyLabeling robot training data →Stereo product labeling data is first-person video from two calibrated cameras of a person applying product and barcode labels to units or cartons by hand or with a label gun on a manufacturing line, with depth and hand pose alongside.
• Task familyThermostat installation robot training data →Stereo thermostat and control wiring data is first-person video from two calibrated cameras of a person mounting a thermostat, connecting control wires to terminals and testing the heating and cooling calls on a trade service call, with depth and hand pose alongside.
• Task familyDrilling robot training data →Stereo drilling and mounting data is first-person video from two calibrated cameras of a person drilling a pilot hole and countersinking, or leveling, marking and screw-mounting a fixture, at a trades workshop bench, with depth and hand pose alongside.
• Task familyPet grooming robot training data →Stereo pet grooming data is first-person video from two calibrated cameras of a person washing, drying, trimming, clipping and brushing animals in a commercial grooming setting, with depth and hand pose alongside.
• Task familyNapkin folding robot training data →Egocentric napkin and towel folding data is real-world video of a person folding napkins to a set style and folding towels and linens, stacked by type, captured with depth, 3D hand pose and frame-accurate action segments on a written spec.
• Task familyPaper trimming robot training data →Egocentric paper trimming and collating data is real-world, first-person video of a person aligning and trimming paper on a trimmer, and collating multi-page sets and stapling each, captured with depth, 3D hand pose and frame-accurate action segments on a written spec.
• Task familyHarvesting robot training data →Stereo crop harvesting data is first-person video from two calibrated cameras of a person picking ripe produce by hand or with shears and placing it in crates, with depth and hand pose alongside.
• Task familyHotel amenity restocking robot training data →Egocentric hotel amenity restocking data is first-person video of a person restocking toiletries, towels, coffee and minibar items from a housekeeping cart, captured by a paid, consented contributor against a written skill spec and delivered as validated episodes with hand pose and action segments.
• Task familyLab automation robot training data →Stereo lab pipetting and sample handling data is first-person video from two calibrated cameras of a person pipetting, transferring liquids and handling sample tubes at a lab bench, with depth and hand pose alongside.
• Task familyGym robot training data →Stereo gym equipment and weight rack data is first-person video from two calibrated cameras of a person setting up benches, racks and machines and returning dumbbells, plates and bars to the weight racks, with depth and hand pose alongside.
• Task familyPottery robot training data →Stereo pottery and ceramics data is first-person video from two calibrated cameras of a person throwing on the wheel, hand-building, sculpting, glazing and loading a kiln, with depth and hand pose alongside.
• Task familyHerb bunching robot training data →Egocentric herb bunching and banding data is video of a person gathering herbs into a bunch and banding it, collected by a paid, consented contributor against a written skill spec and delivered as validated episodes.
• Task familyPlant staking robot training data →Egocentric plant staking and deadheading data is video of a person staking and clipping a stem for support and trimming spent growth, collected by a paid, consented contributor against a written skill spec and delivered as validated episodes.
• Task familyJewelry robot training data →Stereo jewelry and metalsmithing data is first-person video from two calibrated cameras of a person engraving metal, setting a stone, soldering a joint with a torch, wrapping wire and polishing a piece, with depth and hand pose alongside.
• Task familyLeathercraft robot training data →Stereo leathercraft data is first-person video from two calibrated cameras of a person hand-stitching leather, making a bag or wallet, tooling a pattern and working saddlery, with depth and hand pose alongside.
• Task familyVeterinary robot training data →Stereo veterinary clinic data is first-person video from two calibrated cameras of a person preparing an examination room, setting out surgical instruments, cleaning a kennel, handling an animal and giving post-procedure care, with depth and hand pose alongside.
• Task familyWood carving robot training data →Stereo woodcraft and carving data is first-person video from two calibrated cameras of a person crafting furniture, cutting marquetry, whittling, carving wood and turning a blank on a lathe, with depth and hand pose alongside.
• Task familyEmbroidery robot training data →Stereo embroidery and tailoring data is first-person video from two calibrated cameras of a person doing studio textile work, such as hand embroidery, needlepoint, hemming, quilting and tailoring, with depth and hand pose alongside.
• Task familyQuilting robot training data →Egocentric quilting and needlepoint data is first-person video of a person piecing, layering and quilting fabric, or stitching yarn through canvas in a counted pattern, captured by a paid, consented contributor to a written skill spec and delivered as validated episodes with synchronized streams, hand pose and action segments.
• Task familyHanging basket planting robot training data →Egocentric hanging basket planting data is video of a person filling a basket, planting it and hanging it, collected by a paid, consented contributor against a written skill spec and delivered as validated episodes.
• Task familyHair cutting robot training data →Stereo hair salon and barbershop data is first-person video from two calibrated cameras of a person cutting, styling, coloring, washing and blow-drying hair, treating nails and sanitizing a station, with depth and hand pose alongside.
• Task familyDental lab robot training data →Stereo dental lab and instrument reprocessing data is first-person video from two calibrated cameras of a person fabricating dental prostheses or handling, cleaning, sterilizing and assembling surgical instrument sets, with depth and hand pose alongside.
• Task familyEvent venue setup robot training data →Stereo event venue setup and AV cable taping data is first-person video from two calibrated cameras of a person setting up and tearing down staging, routing and taping AV cables, arranging tables and chairs and sweeping a venue, with depth and hand pose alongside.
• Task familyEgocentric lab tube handling data →Egocentric sample tube capping and centrifuge loading data is first-person video of a person filling, capping, vortex-mixing, racking and labeling sample tubes and loading tubes balanced in a centrifuge, captured to a written skill spec and delivered as validated episodes with synchronized streams, hand pose and labeled action segments.
• Task familyBicycle repair robot training data →Stereo bicycle and appliance repair data is first-person video from two calibrated cameras of a person repairing a bicycle or an appliance, with depth and hand pose alongside, so the part, the tool and the hands on it are recorded in three dimensions.
• Task familyWatch repair robot training data →Stereo watch and shoe repair data is first-person video from two calibrated cameras of a person repairing a watch or a shoe, with depth and hand pose alongside, so the small part, the tool and the hands on it are recorded in three dimensions.
• Task familyVacuuming robot training data →Stereo vacuuming and surface sanitizing data is first-person video from two calibrated cameras of a person vacuuming floors and spraying and wiping high-touch surfaces in a commercial building, with depth and hand pose alongside, so the floor, the tool, the surface and the hands on them are recorded in three dimensions.
• Task familyEgocentric lace threading data →Egocentric shoelace and eyelet threading data is first-person video of a person threading a lace through eyelets, as one catalog row, "Thread lace eyelets", states it, captured to a written skill spec and delivered as validated episodes with synchronized streams, 3D hand pose and labeled action segments.
• Task familyEgocentric heatsink mounting data →Egocentric thermal paste and heatsink application data is first-person video of a person applying thermal paste and seating a heatsink, as one catalog row, "Apply paste + heatsink", states it, captured to a written skill spec and delivered as validated episodes with synchronized streams, 3D hand pose and labeled action segments.
• Task familyTractor operation robot training data →Stereo tractor and farm equipment operation data is first-person video from two calibrated cameras of a person operating tractors, mowers or loaders, attaching and detaching implements and working them across the field, with depth and hand pose alongside.
• Task familyEgocentric busbar assembly data →Egocentric busbar and limit block assembly data is first-person video of a person placing a busbar and limit block, driving the screw sequence and removing the block, as one catalog row, "Busbar + limit-block + screws", states it, captured to a written skill spec and delivered as validated episodes with synchronized streams, 3D hand pose and labeled action segments.
• Task familyProduction line inspection robot training data →Stereo production line quality inspection data is first-person video from two calibrated cameras of a person checking parts on a production line against a spec, measuring them and setting aside or rejecting the ones that fail, with depth and hand pose alongside.
• Task rowEgocentric glue-up data →Egocentric clamp and glue joint data is custom first-person video of a paid, consented contributor gluing and clamping a joint square at a workshop bench, captured to a written skill spec and delivered in your training format with consent and anonymization records and a commercial training license.
• Task familyPress-fit assembly robot training data →Stereo press-fit and snap-fit assembly data is first-person video from two calibrated cameras of a person pressing connectors onto a PCB, snapping housing halves together and inserting pegs into holes, with depth and hand pose alongside.
• Task familyBarista robot training data →Stereo barista and espresso data is first-person video from two calibrated cameras of a person working a coffee bar: grinding beans, dosing and tamping, pulling espresso shots, steaming milk and pouring drinks, with depth and hand pose alongside.
• Task familyEgocentric duvet cover data →Egocentric duvet cover insertion data is first-person video and synchronized sensor data of a person putting a duvet into its cover, captured to a written skill spec.
• Task familyKnitting robot training data →Stereo knitting and crochet data is first-person video from two calibrated cameras of a person doing hand and machine knitting or crochet work on garments and textile products, with depth and hand pose alongside.
• Task familyEgocentric sock matching and pairing data →Egocentric sock matching and pairing data is first-person video and synchronized sensor data of a person matching socks and pairing them, captured to a written skill spec.
• Task familyArt studio robot training data →Stereo art studio painting and sculpting data is first-person video from two calibrated cameras of a person painting on canvas, painting a mural, doing printmaking or sculpting, with depth and hand pose alongside.
• Task familyEgocentric mural painting data →Egocentric mural painting data is first-person video of a person sketching an outline on a wall and painting large areas and details with rollers and brushes, recorded by a head-mounted phone or a multiview rig and collected to a written spec.
• Task familyGrocery checkout robot training data →Stereo grocery checkout and bagging data is first-person video from two calibrated cameras of a person scanning items, bagging groceries and handing them to the customer, with depth and hand pose alongside, so the item, the bag and the hands on them are recorded in three dimensions.
• Task familySecurity checkpoint robot training data →Stereo security checkpoint screening data is first-person video from two calibrated cameras of staged checkpoint hand work, such as loading screening trays, loading bags to a belt, searching and re-packing a bag and retrieving a flagged item from a bin, with depth and hand pose alongside.
• Task familyEgocentric poly-bag packing data →Egocentric poly-bagging and multi-item bundling data is first-person video of a person bagging an item and applying a tag, kitting and bundling a multi-item order, and weighing, verifying and sealing a package, recorded with depth and 3D hand pose alongside.
• Task familyCatering robot training data →Stereo catering and banquet plating data is first-person video from two calibrated cameras of catering kitchen hand work, such as portioning hot dishes onto plates in sequence, prepping food in large batches and setting out and clearing a service area, with depth and hand pose alongside.
• Task familyEgocentric depalletizing data →Egocentric depalletizing and cross-dock relabeling data is first-person video of a person removing goods from a pallet and shelving them, stripping and re-labeling for a re-route, and applying and aligning shipping labels, recorded with depth and 3D hand pose alongside.
• Task familyKitchen cleaning robot training data →Stereo kitchen deep clean and degreasing data is first-person video from two calibrated cameras of a person spraying degreaser, scrubbing and rinsing hoods, walls and counters, and emptying and scrubbing fryers, hoods, floors and shelving in a commercial kitchen, with depth and hand pose alongside.
• Task familyEgocentric kitchen degreasing data →Egocentric kitchen deep clean and degreasing data is first-person video of a person spraying degreaser, scrubbing and rinsing hoods, walls and counters, and emptying and scrubbing fryers, hoods, floors and shelving in a commercial kitchen, captured by a paid, consented contributor to a written skill spec.
• Task familyRetail merchandising robot training data →Stereo fitting room and merchandising reset data is first-person video from two calibrated cameras of a person collecting and rehanging garments left in fitting rooms, sorting a rail by size and style, and arranging and facing products on shelves and displays, with depth and hand pose alongside.
• Task familyOffice robot training data →Stereo office mailroom and document handling data is first-person video from two calibrated cameras of a person opening, sorting and distributing mail, filing documents in cabinets, feeding pages into a scanner and loading and collecting from a printer, with depth and hand pose alongside.
• Task familyEgocentric server hardware data →Egocentric server hot-swap and component seating data is real-world video of a person swapping or seating a part in a server, such as a drive caddy, a power supply, a fan, memory modules or an add-in card, recorded to a written spec.
• Task familyEgocentric network cabling data →Egocentric network patching and fiber transceiver data is real-world video of a person patching Ethernet cabling and seating a transceiver with an LC fiber at a server rack, recorded to a written spec.
• Task familyGreenhouse robot training data →Stereo greenhouse potting and pruning data is first-person video from two calibrated cameras of a person potting plants, pruning with shears, seeding trays, transplanting seedlings, staking and clipping stems, training stems on a trellis and trimming spent growth on a greenhouse bench, with depth and hand pose alongside.
• Task familyPharmacy robot training data →Stereo pharmacy medication dispensing and labeling data is first-person video from two calibrated cameras of a person reading a prescription, counting or measuring medication, packaging and labeling it, and printing and applying labels to tubes, bottles and dispensed medication, with depth and hand pose alongside.
• Task familyHedge trimming robot training data →Stereo hedge trimming and pruning data is first-person video from two calibrated cameras of a person cutting hedge tops and sides with trimmers, removing thick stems with loppers, clearing the cuttings and cutting back dead and overgrown branches, with depth and hand pose alongside.
• Task familyPool maintenance robot training data →Stereo pool and spa area maintenance data is first-person video from two calibrated cameras of a person skimming and testing a hotel pool, arranging loungers, wiping surfaces and restocking towels, with depth and hand pose alongside.
• Task familyEgocentric hospital room turnover data →Egocentric patient room prep and surface decontamination data is first-person video of a person making the bed, stocking supplies, positioning equipment and wiping down beds, rails and equipment, with 3D hand pose, depth and frame-accurate action segments.
• Task familyCar battery replacement robot training data →Stereo car battery test and replacement data is first-person video from two calibrated cameras of a person testing a car battery with a multimeter, unbolting and swapping it, and cleaning or tightening the terminals, with depth and hand pose alongside.

Need a specific market? See egocentric data collection by country. Need a specific workplace? Browse Western commercial environments and tasks.

Point the pipeline at a single modality — audio & speech, image, video, text, or multimodal. See it run worldwide on custom collection across 150+ countries, or start from our core specialism, embodied AI training data. For robots and other embodied systems specifically, see physical AI training data or video data collection for robotics. For a dense, high-diversity pretraining domain, see kitchen task pretraining data, or for first-person food prep specifically, egocentric cooking video data collection, or its knife-work counterpart, egocentric cutting and food prep video data, for kneading dough and mixing batter, egocentric baking and dough handling video data, for stacking, carrying and palletizing boxes and parcels, egocentric box and parcel handling video data, for loading trailers and unloading trucks at the dock, egocentric truck loading and unloading video data, for pulling, bagging and carrying out trash and recycling, egocentric trash and recycling task video data, for making coffee and bar drinks, egocentric beverage and coffee prep video data, for taking groceries out of bags and putting them away, egocentric grocery put-away video data, for dismounting, mounting and balancing tires and wheels, egocentric tire and wheel service video data, for watering, potting, pruning and planting, egocentric plant care and gardening video data, for mowing, trimming, planting and hedge pruning on outdoor grounds, egocentric landscaping and groundskeeping video data, for bench and rack wiring work, egocentric cable routing and wiring video data, for mounting a thermostat, connecting control wires to terminals and testing the heating and cooling calls, egocentric thermostat and control wiring data, for server rack and IT hardware bench tasks, egocentric data center and IT hardware task video data, for swapping drive, power supply and fan modules and seating memory and add-in cards, egocentric server hot-swap and component seating data, for patching Ethernet cabling and seating a transceiver with an LC fiber, egocentric network patching and fiber transceiver data, for end-of-line cartoning, sealing, labeling, palletizing and wrap-and-fold, egocentric print and packaging line video data, for placing components and seating connectors on boards, egocentric electronics assembly video data, for scanning, label printing and package labeling, egocentric barcode scanning and labeling video data, for picking items to totes with a handheld scanner in stereo and MCAP, stereo warehouse order picking data, for unscrewing devices, removing boards and batteries and sorting materials in stereo and MCAP, stereo e-waste disassembly and sorting data, for diagnosing and repairing phones, laptops, tablets and consumer electronics in stereo and MCAP, stereo electronics and device repair data, for scanning shelf locations and counting stock in stereo and MCAP, stereo inventory cycle counting data, for jacking cars, replacing brake pads and rotors and torquing bolts in stereo and MCAP, stereo brake pad and rotor service data, for draining engine oil, replacing oil filters and topping up fluids in stereo and MCAP, stereo oil and fluid change data, for hosing down equipment, foaming, scrubbing and rinsing a food plant production area in stereo and MCAP, stereo food plant sanitation washdown data, for grinding weld seams flush with an angle grinder and finishing edges with files in stereo and MCAP, stereo grinding and deburring data, for testing a car battery with a multimeter, swapping it and tightening the terminals in stereo and MCAP, stereo car battery test and replacement data, for testing a car battery with a multimeter, swapping it and working the terminals, egocentric car battery replacement data, for connecting a scan tool, reading fault codes and inspecting components, egocentric vehicle diagnostics data, for inspecting, removing and installing air and cabin filters with a screwdriver and a shop rag, egocentric air and cabin filter replacement data, for vacuuming carpets, hard floors and vehicle interiors, egocentric vacuuming data, for spraying disinfectant, scrubbing toilets, urinals and sinks, mopping and restocking in a restroom in stereo and MCAP, stereo restroom cleaning and disinfection data, for pre-spraying carpets, running an extractor in passes and positioning air movers in stereo and MCAP, stereo carpet extraction data, for receiving inbound pallets with a pallet jack, unloading trucks, moving freight with a hand truck and loading outbound trailers in stereo and MCAP, stereo warehouse receiving and truck loading data, for printing labels, applying them to outbound boxes and sorting boxes by carrier, route or type in stereo and MCAP, stereo warehouse sorting and labeling data, for grinding, tamping, pulling espresso shots and steaming milk, egocentric barista and espresso data, for the same coffee bar work in stereo and MCAP, stereo barista and espresso data, for wiping counters and cleaning surfaces, egocentric wiping and surface cleaning video data, for mopping, sweeping and vacuuming floors, egocentric floor cleaning and mopping video data, for scrubbing toilets, sinks and mirrors in bathrooms and restrooms, egocentric bathroom cleaning video data, for resetting a hotel guest room after checkout, egocentric hotel room turnover data, for stripping and making beds, vacuuming and dusting in stereo and MCAP, stereo hotel room turnover and housekeeping data, for restocking toiletries, towels, coffee and minibar items from the housekeeping cart, egocentric hotel amenity restocking data, for rolling, brushing and finishing surfaces, egocentric painting and coating video data, for cutting, sanding, drilling and joining wood, egocentric woodworking and carpentry video data, for carving and whittling wood by hand, egocentric wood carving video data, for cutting veneer pieces to pattern and gluing them into a decorative inlay on a panel, egocentric marquetry and inlay data, for putting flat-pack and workshop furniture together, egocentric furniture assembly video data, for plating dishes, portioning food and serving to a table, egocentric meal plating and serving video data, for counter service, fryer and grill work and order assembly, egocentric fast food and quick-service kitchen video data, for cutting, fitting and installing pipe and fixtures, egocentric plumbing and pipe fitting video data, for counting stock and reconciling totals, egocentric inventory counting and stock check video data, for grooming and caring for animals, egocentric pet care video data, for cutting, coloring and styling hair and nail and skin-care work, egocentric hair salon and beauty task video data, for mixing and applying hair colour with a brush and foils, egocentric hair coloring data, for threading laces and coiling and strapping cables, egocentric knot tying and rope handling data, for stacking, wrapping and breaking down pallets, egocentric pallet building and wrapping video data, for washing, drying and polishing vehicles, egocentric car washing and vehicle detailing video data, for oil changes, filters, batteries, brakes and diagnostics, egocentric vehicle maintenance video data, for welding, grinding and fitting metal parts, egocentric welding and metal fabrication video data, for diagnosing, opening and replacing parts in phones and laptops, egocentric electronics and device repair video data, for opening a watch case, replacing the battery or movement parts and resealing, egocentric watch and small mechanism repair data, for carrying parcels to the door and handing them to the recipient, egocentric last-mile package delivery video data, for collecting a food order at the restaurant, bagging it in an insulated carrier and delivering it, egocentric food delivery pickup and handoff data, for stripping an old finish, repairing joints and veneer, then sanding and refinishing a piece of furniture, stereo furniture restoration and refinishing data, for throwing, hand-building and glazing clay, egocentric pottery and ceramics video data, for engraving, setting stones, soldering, wire wrapping and polishing metal, egocentric jewelry making and metalsmithing video data, for seating a stone and cutting letters and patterns into metal, egocentric gem setting and engraving video data, for sorting, weighing and packing fruit and grading crops, egocentric produce packing and grading video data, for sorting, weighing, packing and labeling strawberries and grading produce in stereo and MCAP, stereo produce sorting and packing data, for inspecting, filling, sealing, labelling and washing down in a food or beverage plant, egocentric food and beverage manufacturing video data, for banquet plating, bulk prep and event service setup in a catering kitchen, egocentric catering and event kitchen video data, for banquet plating, bulk prep and service setup on a stereo pair delivered as MCAP, stereo catering and banquet plating data, for equipment setup, floor cleaning and weight rack organization in a gym, egocentric gym and fitness facility task video data, for tables, chairs, staging and cables at an event venue, egocentric event venue setup and teardown video data, for tray loading, bag placement and item retrieval on a staged checkpoint, egocentric security checkpoint task video data, for hotel luggage delivery and checkpoint bag loading, tag scan and re-pack, egocentric luggage and baggage handling data, for pool skimming and testing, lounger setup, wiping and towels in a hotel pool area, egocentric pool and spa maintenance data, for fastening buttons and snaps, running a zipper, threading lace eyelets and strapping a cable on a cloth bench, egocentric buttons, zippers and fasteners data, for working the levers and control panels of presses, cranes or machinery from the operator station, egocentric heavy machinery operation data, for handling, cutting, forming and finishing glass, egocentric glass manufacturing data, for sorting stock onto racks and bins and sorting damaged goods in a store backroom, egocentric retail backroom organization data, for stereo capture of rolling, cutting in, priming and staining with depth and hand pose, delivered as MCAP, stereo painting and coating data, for stereo capture of spreading adhesive, setting tiles and grouting joints with depth and hand pose, delivered as MCAP, stereo tile and flooring installation data, for stereo capture of hanging, taping and mudding drywall with depth and hand pose, delivered as MCAP, stereo drywall installation and finishing data, for examination room prep, instrument setup and kennel cleaning in a veterinary clinic, egocentric veterinary clinic task video data, for stereo capture of the same clinic tasks with depth and hand pose, delivered as MCAP, stereo veterinary clinic data, for worn tactile gloves with hand tracking, egocentric tactile glove and hand tracking video data, for work recorded by people doing their actual jobs, egocentric video data from real workers doing real jobs, for how accepted data reaches your bucket, egocentric data delivery, transfer and storage, and for first-person cloth and garment handling, laundry folding video data for robot training, for pressing, hanging and sorting garments, egocentric ironing and garment handling video data, for cutting, hemming and stitching fabric, egocentric sewing and textile handling video data, for stitching decorative patterns and yarn into cloth by hand, egocentric embroidery and hand stitching data, for punching, stitching, tooling and fitting leather, egocentric leathercraft video data, for making beds and changing linens, egocentric bed making video data, for putting a duvet into its cover, egocentric duvet cover insertion data, for matching socks and pairing them, egocentric sock matching and pairing data, and for first-person office and desk work, egocentric video data of office and desk tasks, for printer trays, paper jams and shredding, egocentric office printer and shredder data, for stuffing and sealing envelopes and packing and taping mailers, egocentric envelope stuffing and mailer packing data, for cutting and fitting steel pipe in a metal workshop, stereo pipe cutting and fitting data, for inspecting and cleaning up welds in a metal workshop, stereo weld inspection and cleanup data, for inspecting and sorting damaged goods in a retail back of house, egocentric damaged goods sorting data, for replacing a water heater on a trade service call, stereo water heater replacement data, for snaking and clearing a drain on a service call, stereo drain snaking and clearing data, for running, ramping and taping AV cables at an event venue, egocentric AV cable management and taping data, for trimming flash from molded parts with flush cutters and a deburring blade, stereo part trimming and deflashing data, for applying product and barcode labels to units or cartons by hand or with a label gun, stereo product labeling data, for mounting a thermostat, connecting control wires to terminals and testing the heating and cooling calls in stereo, stereo thermostat and control wiring data, for folding napkins to a set style and folding towels and linens on a hospitality service set, egocentric napkin and towel folding data, for drilling a pilot hole, countersinking and screw-mounting a fixture at a trades workshop bench in stereo, stereo drilling and mounting data, for bathing, drying, trimming and brushing animals in a commercial grooming salon in stereo, stereo pet grooming data, for aligning and trimming paper on a trimmer and collating and stapling multi-page sets at an office document station, egocentric paper trimming and collating data, for gathering herbs into bunches and banding them on a greenhouse grow-bench set, egocentric herb bunching and banding data, for staking and clipping plant stems and trimming and deadheading spent growth on a greenhouse grow-bench set, egocentric plant staking and deadheading data, for filling, planting and hanging a basket on a greenhouse grow-bench set, egocentric hanging basket planting data, for picking ripe produce, cutting it with pruning shears and placing it in field crates in stereo, stereo crop harvesting data, for operating tractors, mowers and loaders and hitching implements in stereo, stereo tractor and farm equipment operation data, for pipetting, liquid handling and sample tubes at a lab bench in stereo, stereo lab pipetting and sample handling data, for capping, racking and labeling sample tubes and loading a centrifuge as first-person video, egocentric sample tube capping and centrifuge loading data, for threading a lace through eyelets as first-person video, egocentric shoelace and eyelet threading data, for applying thermal paste and seating a heatsink as first-person video, egocentric thermal paste and heatsink application data, for placing a busbar and limit block and driving the screw sequence as first-person video, egocentric busbar and limit block assembly data, for setting up gym equipment and returning weights to the racks in stereo, stereo gym equipment and weight rack data, for throwing, hand-building, glazing and kiln loading in a pottery studio in stereo, stereo pottery and ceramics data, for gem setting, engraving, torch soldering and polishing at a jeweler bench in stereo, stereo jewelry and metalsmithing data, for saddle stitching, tooling and saddlery at a leather bench in stereo, stereo leathercraft data, for wood carving, whittling, woodturning, marquetry and furniture crafting at a studio bench in stereo, stereo woodcraft and carving data, for hand embroidery, needlepoint, hemming, tailoring and quilting at a studio bench in stereo, stereo embroidery and tailoring data, for hand and machine knitting and crochet of garments and textile products in stereo, stereo knitting and crochet data, for canvas painting, mural work, printmaking and sculpting in an art studio in stereo, stereo art studio painting and sculpting data, for scanning items and bagging groceries at a checkout in stereo, stereo grocery checkout and bagging data, for loading screening trays, bags and bins on a staged mock security checkpoint in stereo, stereo security checkpoint screening data, for bagging and tagging an item, bundling a multi-item order and weighing and sealing a package in first person, egocentric poly-bagging and multi-item bundling data, for removing goods from a pallet to a shelf and stripping and re-labeling for a re-route in first person, egocentric depalletizing and cross-dock relabeling data, for sketching and painting large wall areas and details with rollers and brushes in first person, egocentric mural painting data, for piecing, layering and quilting fabric and counted-stitch needlepoint in first person, egocentric quilting and needlepoint data, for hair cutting, styling, coloring, blow-drying, nail treatments and station sanitation in a salon or barbershop in stereo, stereo hair salon and barbershop data, for dental prosthesis fabrication and surgical instrument handling, cleaning, sterilizing and set assembly in stereo, stereo dental lab and instrument reprocessing data, for event venue staging setup and teardown, AV cable routing and taping, table and chair layout and venue sweeping in stereo, stereo event venue setup and AV cable taping data, for staging platform and stage deck setup and teardown with equipment and lighting in stereo, stereo event staging setup and teardown data, for bicycle repair (brakes, gears, tubes, chains, cables, wheel truing) and appliance repair (opening, component testing, part replacement, reassembly) in stereo, stereo bicycle and appliance repair data, for watch repair (opening the case, replacing the battery or movement parts under magnification, resealing) and shoe repair (soles or heels, replacements, stitching, edges) in stereo, stereo watch and shoe repair data, for commercial vacuuming (carpets and hard floors in offices and corridors, moving chairs) and surface sanitizing (spraying and wiping desks, handles and high-touch surfaces) in stereo, stereo vacuuming and surface sanitizing data, for vehicle interior vacuuming, wiping and conditioning in stereo, stereo car interior vacuuming and detailing data, for cleaning and dressing wheels and tires as first-person capture, egocentric wheel and tire dressing data, for commercial kitchen degreasing (hoods, walls and counters) and deep clean (fryers, hoods, floors and shelving) in stereo, stereo kitchen deep clean and degreasing data, for the same two kitchen rows as first-person capture, egocentric kitchen deep clean and degreasing data, for retail fitting room resets (collecting and rehanging garments), garment rail sizing and merchandising and facing in stereo, stereo fitting room and merchandising reset data, for office mail handling, filing, scanning and printing in stereo, stereo office mailroom and document handling data, for greenhouse potting, pruning, seeding and transplanting in stereo, stereo greenhouse potting and pruning data, for pharmacy medication dispensing and labeling in stereo, stereo pharmacy medication dispensing and labeling data, for quick-service order assembly, wrapping, bagging and counter hand-over in stereo, stereo quick-service order assembly and counter data, and for teams pairing first-person video with language, egocentric video data for vision-language model training. The labels behind that video are covered in egocentric video annotation, and the footage those labels are built from is human activity video data. Every batch ships through the anonymization pass, and the consent and licensing basis is documented in provenance. Not sure custom collection is the right call? Compare it against off-the-shelf datasets.

What is end-to-end custom data collection?

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 — discovery, spec, sourcing, capture, annotation, QA, and licensed delivery — so you brief one team instead of stitching together crowds, tools, and reviewers yourself.

What does "custom collection for AI" actually deliver?

A dataset built to your specification rather than pulled off a shelf — the exact modality, languages, demographics, devices, environments, and edge cases your model needs, collected by paid contributors with signed consent, structured into your delivery format, and handed over under a perpetual, buyer-owned commercial license with a visible reject log and a data card.

Why buy the whole pipeline instead of assembling it yourself?

Because the failures in AI data live in the handoffs. A crowd platform, a labelling vendor, and a QA tool each optimise their own step; the spec and the consent chain get lost between them. End-to-end means one team owns sourcing, capture, annotation, QA, and licensing, so the accept criteria are enforced the whole way through and there is one owner accountable for the result.

Do I need a finished spec before I start?

No. Most projects begin from a gap — a model that fails in a certain language, region, or condition — not a schema. Turning that problem into a written spec with explicit accept and reject criteria is the first stage of the work, and you approve it before any collection begins.

Which types of AI can you collect data for?

LLMs and assistants, speech and ASR, computer vision, embodied AI and robotics, multimodal and sensor models, and safety or evaluation sets. Egocentric first-person video is our core specialism; the other modalities run through the same spec-first, consent-documented pipeline.