Robot learning dataset
RoboMIND
RoboMIND’s current Hugging Face release includes teleoperated robot demonstrations and digital-twin simulation data, stored as HDF5 trajectories with visual observations, robot states and language instructions. The project reports 107k real-world demonstrations for versions 1.1/1.2; the initial version 1.0 was smaller. The official dataset card is tagged Apache-2.0.
Records last verified Oct 13, 2026 · Send a correction
Licence & use
Apache-2.0
Commercial use allowed
Licence text ↗Terms apply to the checked release. Read publisher conditions for mixed collections, converted formats and gated access.
- Publisher
- RoboMIND Team (X-Humanoid, Peking University and Beijing Academy of Artificial Intelligence)
- Released
- Not dated
- Published size
- 107,000 trajectories · 12.3 TB
- Modalities
- rgb, proprioception, language
- Formats
- HDF5
- Real / simulated
- Mixed
- Includes humanoid recordings
- Yes
Size & release notes
Published 107k figure describes real-world demonstrations in versions 1.1/1.2, not the total including simulation. Version 1.0 had 55K trajectories and 279 tasks. Version 1.2 adds one real and nine simulated Upright_Cup tasks. Hugging Face lists 12.3 TB total repository files. Release dates for these versions are not stated.
Recorded embodiments
- Franka Emika Panda
- UR5e
- AgileX Cobot Magic V2.0
- Tien Kung humanoid (model not stated)
Models disclosing this training dataset
More robot learning datasets
Evidence
Sources
- [s1] RoboMIND project page ↗
X-Humanoid · official · accessed Oct 13, 2026
- [s2] RoboMIND: Benchmark on Multi-embodiment Intelligence Normative Data for Robot Manipulation ↗
arXiv · paper · published Dec 18, 2024 · accessed Oct 13, 2026
- [s3] Official RoboMIND dataset card: license, versions and formats ↗
RoboMIND Team · official · accessed Oct 13, 2026
- [s4] Apache License, Version 2.0 ↗
Apache Software Foundation · official · accessed Oct 13, 2026