Apache 2.0
The publisher's licence permits commercial use and redistribution, with attribution. Hub card tag apache-2.0 with a badge linking to the Apache 2.0 text; no licence sentence in the card body. The gate conditions were not visible without login.
The host asks you to sign in to Hugging Face and accept the publisher's terms before it serves files.
hf download x-humanoid-robomind/RoboMIND --repo-type dataset --local-dir ./RoboMINDContact-information gate; sign in first. 12.3 TB on xet storage. Utilities at github.com/Open-X-Humanoid/RoboMIND-dataset-utils; quick_start.ipynb in the repo.
Specification, as published
- Episodes (stated)
- 107,000 episodes
- Scale
- 107k real-world trajectories: 52,926 Franka, 25,170 UR5e, 19,152 Tien Kung, 10,629 AgileX
- Task
- Pick & place, Sorting & packing, Cooking & food prep, Tidying & put-away, Other
- Environment
- Indoor – home, Indoor – office, Indoor – lab, Indoor – warehouse
- Modality
- RGB, Depth, Proprioception, End-effector pose, Language
- Capture
- Egocentric
- In frame
- Robot hands
- Captured in
- China
- Embodiment
- Franka Emika Panda, UR5e, AgileX Cobot Magic V2.0 dual-arm, Tien Kung humanoid with dexterous hands
- Formats
- HDF5, CSV, JSON
- Size
- 12.3 TB
- Hosted on
- Hugging Face
Figures are the publisher's own. Kinetic Blocks has not measured this dataset.
About this dataset
RoboMIND (Multi-embodiment Intelligence Normative Data) is a human-teleoperation benchmark across four embodiments: Franka Emika Panda, UR5e, AgileX Cobot Magic V2.0 dual-arm and the Tien Kung humanoid with dual dexterous hands. It covers 479 tasks in six categories over 96 object classes in domestic, industrial, kitchen, office and retail settings, plus 5k real-world failure demonstrations with causes, and includes an Isaac Sim digital-twin subset that should be excluded from any real-footage count. Streams are multi-view RGB (one or three cameras by embodiment), depth where available, proprioceptive state, end-effector poses via URDF scripts, language instructions and frame-level annotations on 10,000 trajectories, stored as one HDF5 per episode with CSV instructions and JSON annotations. The card states no hours or frame rate. A V2.0 collection is announced on ModelScope.
Publisher's own task names: 479 tasks in six categories, 96 object classes, 5k failure demonstrations with causes.
What Kinetic Blocks did, and did not do
Indexed and linked from the publisher. Not hosted, verified or graded by Kinetic Blocks. The publisher's terms govern. The licence shown is the one stated on the publisher's page, linked above. Nothing on this page is a claim by Kinetic Blocks.
Want the verified ones too?
Request access to the marketplace, where every paid listing is verified against its files and graded before it goes on sale.