# DataClue > DataClue licenses real-world AI training data. In stock today: 800,000+ hours of egocentric (first-person) video of people doing real work, for robot learning, embodied AI, VLA pretraining and world models. Buyers can preview sample clips, see measured capture specs and request a quote online. Delivery: original files at 1600×1200 or 1920×1080 depending on device and session; preview clips on the site are lower-resolution demos. Capture: monocular GoPro HERO12/13 with IMU, smart glasses, and an in-house stereo rig (also for sale). UMI data: 1,320,332 UMI (Universal Manipulation Interface) recordings of household manipulation, 70,680 valid hours, six synchronised data streams per recording, subtask segmentation, semantic action labels and start/end timestamps on all data. Details: https://dataclue.app/company Custom collection: raw video capture or annotated video (including hand annotation) to your spec. Pricing: $40/hour for complex scenarios (tool use, deformables, precision assembly) and $30/hour for standard scenarios, under a non-exclusive commercial training licence. Prices include full annotation: action segmentation, step-level narration and 2D hand keypoints (3D hand pose on stereo sessions). Exclusive licences, custom label specs and custom collection are quoted per project. Payment is by invoice after the licence agreement is signed. Every recording is made with the written consent of the people filmed. Consent documentation is available to buyers during licence review. Contact: contact@dataclue.app · Jersey City, United States and Hong Kong · Request a quote: https://dataclue.app/request-quote Free sample (18 de-identified clips with capture specs, evaluation licence): https://huggingface.co/datasets/DataClue/dataclue-egocentric-sample ## Datasets - [Manufacturing & Production](https://dataclue.app/datasets/manufacturing-operations): 176,000 hours in stock, $40/hour. Bench assembly, part sorting and electronic component insertion recorded at the workstation. Scenarios: Screw and Nut Sorting and Assembly; Small Electronic Component Insertion and Assembly; Plastic Housing and Metal Component Assembly. - [Household Services](https://dataclue.app/datasets/household-activities): 152,000 hours in stock, $40/hour. Cooking, food prep and laundry filmed from the wearer's point of view in real homes. Scenarios: Pancake Cooking and Plating; Vegetable Cutting and Preparation; Handwashing Laundry. - [Warehousing & Logistics](https://dataclue.app/datasets/warehouse-logistics): 136,000 hours in stock, $30/hour. Stocking shelves, folding parcel boxes and moving pallets — logistics work at human scale. Scenarios: Stocking Warehouse Shelves; Parcel Box Folding; Manual Pallet Truck Operation. - [Retail Services](https://dataclue.app/datasets/retail-services): 120,000 hours in stock, $30/hour. Supermarket and pharmacy picking recorded from the shopper's and picker's point of view. Scenarios: Product Selection and Shopping Cart Loading; Pharmacy Product Picking and Basket Loading. - [Food Service & Hospitality](https://dataclue.app/datasets/food-service): 104,000 hours in stock, $30/hour. Tableware handling and drinks prep — repetitive, high-volume service tasks in long sessions. Scenarios: Tableware Wiping; Beverage Cup and Lid Assembly. - [Specialized Operations](https://dataclue.app/datasets/specialized-operations): 72,000 hours in stock, $30–40/hour. Skilled trades, office handling and library work — the long tail of real-world tasks. Scenarios: Wall Thermostat Installation; Manual Metal Pipe Assembly; Cable Coiling and Organization; Paper Clip Linking and Arrangement; Moving Packaging Materials in a Shared Area; Library Lending Form and Book Organization. - [Hospitality & Accommodation](https://dataclue.app/datasets/hospitality-accommodation): 40,000 hours in stock, $40/hour. Housekeeping work in guest and care-facility rooms, recorded from the worker's point of view. Scenarios: Bed Making and Linen Change. - [UMI Gripper Demonstrations](https://dataclue.app/datasets/umi-gripper-demonstrations): 70,680 hours in stock, Quoted per project. Handheld-gripper demonstrations of household tasks: six synchronised cameras, gripper trajectories and subtask labels. Scenarios: Washing Cherry Tomatoes; Fitting a Sofa Cover; Clearing and Wiping a Kitchen Counter; Clearing Tableware to the Sink; Putting Away Small Clothes; Rinsing Utensils at the Sink; Shoe Arrangement. ## Categories - [Egocentric video datasets](https://dataclue.app/datasets/egocentric-video): Buy egocentric (first-person) video datasets for robotics and embodied AI: 800,000+ hours across manufacturing, household, warehousing and logistics, retail, food service, specialized operations and hospitality. - [Robot training datasets](https://dataclue.app/datasets/robotics): Robot training datasets for imitation learning and VLA pretraining: real-world egocentric manipulation video across seven environments, plus custom teleoperation collection. - [Multimodal datasets: video + IMU + hand pose](https://dataclue.app/datasets/multimodal): Multimodal training data for embodied AI: egocentric video with embedded IMU telemetry, stereo capture and hand-pose annotation included. - [Teleoperation data — custom collection](https://dataclue.app/datasets/teleoperation): Robot teleoperation data collected to your task spec: joint states, actions and synchronised camera streams. Custom collection with exclusive licensing available. - [Brain-computer interface (BCI) data — custom collection](https://dataclue.app/datasets/bci): Brain-computer interface data collected to your protocol: non-invasive EEG recordings with your task paradigm, participants and device, written consent from every participant, exclusive licensing available. ## Custom collection (not in stock) - Robot Teleoperation Data: Robot-side demonstrations — joint states, actions and wrist/head camera video — collected on your hardware or ours to your task spec. - Brain-Computer Interface (BCI) Data: Non-invasive brain-signal (EEG) recordings collected to your protocol: task paradigm, participants, device and channel count scoped per project, with written consent from every participant. - New Environments & Scenarios: Healthcare support, agriculture, automotive, outdoor work and other settings not yet in stock, recorded to your scenario list. - Custom annotation: Label specs beyond the included package (action segments, narration, hand keypoints): object boxes and masks, your own taxonomy, or extra QA passes on in-stock or custom footage. ## Guides - [Training data for embodied AI](https://dataclue.app/solutions/embodied-ai): Embodied AI training data for robot foundation models, VLA pretraining and dexterous manipulation: 800,000+ hours of real-world egocentric video across seven environments. - [Training data for world models](https://dataclue.app/solutions/world-models): World model training data: long, continuous egocentric video with IMU egomotion across real environments. Licensable per hour with verified sample specs. - [How to buy robotics training data: a buyer's checklist](https://dataclue.app/resources/how-to-buy-robotics-training-data): A practical checklist for buying robotics training data: specs to verify, licence terms to negotiate, inventory questions to ask and how to run a pilot before committing budget. - [What is egocentric video data, and why do robots need it?](https://dataclue.app/resources/what-is-egocentric-video-data): Egocentric (first-person) video explained: how it is captured, why it matters for robot learning and world models, and what to look for in a first-person video dataset. - [Exclusive vs non-exclusive licensing for AI training data](https://dataclue.app/resources/exclusive-vs-non-exclusive-data-licensing): How exclusive and non-exclusive data licences differ for AI training, what each costs, and when exclusivity is worth paying for. - [Licensing](https://dataclue.app/licensing): how DataClue data is licensed and paid for. - [Company & data sourcing](https://dataclue.app/company): data is recorded first-hand on our own devices with written consent; quality control and how samples match delivered data. ## For AI agents (MCP) DataClue runs a public MCP server (streamable HTTP, no auth): https://dataclue.app/mcp Tools: search_datasets, get_dataset, get_pricing_and_terms, request_quote (sends a sample or quote request to sales on the user's behalf; requires the user's work email). Setup: https://dataclue.app/ai-agents ## Buying - Free evaluation samples: https://dataclue.app/request-quote?plan=sample (work email only) - 10-hour pilot: https://dataclue.app/request-quote?plan=pilot&dataset= - Volume quote: https://dataclue.app/request-quote?plan=volume&dataset=&hours=1000 - The end buyer is verified at licence review, before a commercial licence is signed.