Commercial training data from specialist providers. Listings show what each source advertises; licensing is confirmed with the provider before any agreement.
The Multimodal Human Manipulation Dataset is a 6,000-hour sensor-rich collection of humans performing physical manipulation tasks with synchronized visual, motion, and…
A single continuous egocentric demonstration of a person opening a cardboard box with a utility knife and unpacking its contents.
A single continuous egocentric demonstration of a person painting a window trim with a brush on an outdoor building exterior — loading the brush, applying paint along…
Multi-view humanoid teleoperation demonstrations of a contact-rich press-and-release manipulation task, captured on a Unitree G1 robot operated by a human teleoperator.
A single continuous egocentric demonstration of a person setting a round dining table — laying out plates, bowls and cutlery.
A single continuous egocentric demonstration of a person unpacking and assembling a camera tripod — removing it from its bag, extending and locking the legs, and…
A single showcase episode of handheld-gripper tabletop manipulation, converted from raw MCAP into a HuggingFace LeRobot v2.1 dataset and shipped with the complete QC…
A single continuous egocentric demonstration of a person wrapping a gift — handling wrapping paper, tape and scissors on a work surface.
This dataset provides high-quality, ethically sourced Annotated Imagery Data and video streams across Africa and global regions, specifically designed to train,…
Specifications: • Volume: 50,000 hours • Weekly output: 1,000+ hours of new footage added per week • Environments: 16 distinct settings — 1 residential (home & daily…
Specifications: • Volume: 50,000 hours • Weekly output: 1,000+ hours of new footage added per week • Environments: 16 distinct settings — 1 residential (home & daily…
The provider provides commercially licensable multi-view drone imagery and associated reconstructed 3D and geospatial data.
Specifications: • Scope: Any modality — audio & speech, video & image, embodied AI/egocentric, 3D & spatial, generative media, text & multimodal • Sourcing: Proprietary…
Off-the-shelf manipulation datasets rarely match the exact task, objects, or environment your policy needs.
A teleoperated bimanual demonstration of domestic garment handling — picking up and manipulating clothing items over a soft surface — recorded on a DAS bimanual…
A collection of large scale real‐world robotics manipulation dataset collected using the dual‐arm humanoid robot.
Specifications: • Volume: 50,000 hours • Weekly output: 1,000+ hours of new footage added per week • Environments: 16 distinct settings — 1 residential (home & daily…
A collection of first-person videos of various daily activities like cleaning, walk, drive, painting, etc.
Azimov is egocentric (first-person) data recorded by trained professionals doing their actual jobs on live, operating sites.
Egocentric Data Specifications: • Volume: 50,000 hours • Weekly output: 1,000+ hours of new footage added per week • Environments: 16 distinct settings — 1 residential…
Specifications: • Volume: 50,000 hours • Weekly output: 1,000+ hours of new footage added per week • Environments: 16 distinct settings — 1 residential (home & daily…
33 minutes and 4 seconds of continuous, unscripted laundry folding on a bed in a North American home (Utah, USA), recorded head-mounted on a Luxonis OAK-D Pro W on 21…
This dataset contains ~20-second episodes of a person naturally folding cloth, captured by the provider’s wrist-worn AI smartwatch.
A collection of egocentric videos captured using head‐mounted, 0.6× ultrawide perspective cameras (1080p at 30 fps), showing individuals performing a variety of tasks…