Note

This is a Hugging Face dataset. For large datasets, ensure huggingface_hub>=1.1.3 to avoid rate limits. Learn more in the Hugging Face integration docs.

Hugging Face

RoboMIND → FiftyOne (Native Multimodal MCAP)#

preview

A 32-episode subset of x-humanoid-robomind/RoboMIND, eight episodes from each of four robot embodiments (Franka, AgileX, Tien Kung, UR), converted to native multimodal MCAP episodes. Each episode carries per-camera RGB and depth streams, joint telemetry for every recorded arm with timeline plot channels, and the language instruction.

Installation#

pip install fiftyone

Usage#

import fiftyone as fo
import fiftyone.utils.huggingface as fouh

dataset = fouh.load_from_hub(
    "Voxel51/RoboMIND",
    name="RoboMIND",
    persistent=True,
)
fo.launch_app(dataset)

What you get#

  • 32 .mcap episodes across franka, agilex, tienkung, and ur

  • Per-embodiment telemetry channels discovered from the source layout (for example /puppet-joint-position, /master-joint-velocity-left), each with a timeline plot channel

  • Per-episode fields: embodiment, task, episode_id, num_frames, duration

License & attribution#

The source dataset is released under Apache 2.0. Changes from the source: episode subsetting and conversion from HDF5 to MCAP. The source files carry no per-frame timestamps; frames are timed at the 10 Hz cadence used by RoboMIND’s own visualization tooling.

Citation#

@article{wu2024robomind,
  title={RoboMIND: Benchmark on Multi-embodiment Intelligence Normative Data for Robot Manipulation},
  author={Wu, Kun and Hou, Chengkai and Liu, Jiaming and Che, Zhengping and others},
  journal={arXiv preprint arXiv:2412.13877},
  year={2024}
}