Datasets:

Xia-2004
/

red_cube

license: apache-2.0
task_categories:
  - robotics
pretty_name: ARX Red Cube YuHai HDF5 Dataset
size_categories:
  - n<1K
tags:
  - robotics
  - arx
  - teleoperation
  - hdf5
  - visuomotor

ARX Red Cube YuHai HDF5 Dataset

This dataset contains ARX-X5 left-arm single-arm teleoperation episodes for a red cube manipulation task. Each episode is stored as one HDF5 file.

https://huggingface.co/datasets/Xia-2004/red_cube

Files

episode_000000.hdf5 ... episode_000200.hdf5
201 episodes
42,165 total frames
RGB images are stored as uint8
Actions are stored as float32

HDF5 Format

Each episode_*.hdf5 contains:

Key	Shape	Dtype	Meaning
`action`	`(T, 5)`	`float32`	Delta end-effector action for each frame
`timestamp`	`(T,)`	`float64`	Collection timestamp for each frame
`pixels/camera_third`	`(T, 224, 224, 3)`	`uint8`	Third-person RGB camera frames
`goal_pixels/camera_third`	`(224, 224, 3)`	`uint8`	Final goal image from `camera_third`
`observations/pixels/camera_third`	soft link	-	Link to `pixels/camera_third`

The action columns are:

[dx, dy, dz, dyaw, d_gripper]

where:

dx, dy, dz: delta end-effector translation
dyaw: delta end-effector yaw
d_gripper: delta gripper command

The HDF5 attribute action_order also records this column order.

Python Loading Example

import h5py

path = "episode_000000.hdf5"

with h5py.File(path, "r") as f:
    actions = f["action"][:]
    camera_third = f["pixels/camera_third"][:]
    goal_camera_third = f["goal_pixels/camera_third"][:]

print(actions.shape)
print(camera_third.shape)
print(goal_camera_third.shape)

Download

Install the Hugging Face Hub CLI:

pip install -U huggingface_hub

Download the full dataset:

huggingface-cli download Xia-2004/red_cube \
  --repo-type dataset \
  --local-dir ./red_cube

Or with Python:

from huggingface_hub import snapshot_download

snapshot_download(
    repo_id="Xia-2004/red_cube",
    repo_type="dataset",
    local_dir="./red_cube",
)