OpenGalaxea/Galaxea-Open-World-Dataset

🚀 Galaxea Open-World Dataset

Key features

• 500+ hours of real-world mobile manipulation data.
• All data collected using one uniform robotic embodiment for consistency.
• Fine-grained subtask language annotations.
• Covers residential, kitchen, retail, and office settings.
• Dataset in RLDS format.

Dataset Structure

For convenience, we divided the 500 hours of data into four equal parts by time. We also provide a small sample dataset for quick start.

rlds
├── part1_r1_lite
│   ├── 1.0.0
│   │   ├── dataset_info.json
│   │   ├── features.json
│   │   ├── merge_dataset_large_r1_lite-train.tfrecord-00000-of-02048
│   │   ├── ...
│   │   ├── merge_dataset_large_r1_lite-train.tfrecord-02047-of-02048
├── part2_r1_lite
├── part3_r1_lite
├── part4_r1_lite
├── sample
│   ├── 1.0.0
│   │   ├── merge_dataset_large_r1_lite-train.tfrecord-00000-of-01024
│   │   ├── ...
│   │   ├── merge_dataset_large_r1_lite-train.tfrecord-01023-of-01024

Dataset Schema

OpenGalaxeaDataset = {
    "episode_metadata": {
        "file_path": tf.Text,  # path to the original data file
    },
    "steps": {
        "is_first": tf.Scalar(dtype=bool),  # true on first step of the episode
        "is_last": tf.Scalar(dtype=bool),  # true on last step of the episode

        "language_instruction": tf.Text,  # language instruction, format: "high level"@"low level chinese"@"low level english"
        "observation": {
            "base_velocity": tf.Tensor(3, dtype=float32),   # robot base velocity
            "gripper_state_left": tf.Tensor(1, dtype=float32),  # left gripper state, 0-close and 100-open
            "gripper_state_right": tf.Tensor(1, dtype=float32), # right gripper state, 0-close and 100-open
            "depth_camera_wrist_left": tf.Tensor(224, 224, 1, dtype=uint16),  # wrist camera depth left viewpoint, unit: mm
            "depth_camera_wrist_right": tf.Tensor(224, 224, 1, dtype=uint16),  # wrist camera depth right viewpoint, unit: mm
            "image_camera_head": tf.Tensor(224, 224, 3, dtype=uint8), # head camera RGB viewpoint
            "image_camera_wrist_left": tf.Tensor(224, 224, 3, dtype=uint8), # wrist camera RGB left viewpoint
            "image_camera_wrist_right": tf.Tensor(224, 224, 3, dtype=uint8), # wrist camera RGB right viewpoint
            "joint_position_arm_left": tf.Tensor(6, dtype=float32), # joint positions of the left arm
            "joint_position_arm_right": tf.Tensor(6, dtype=float32), # joint positions of the right arm
            "joint_position_torso": tf.Tensor(4, dtype=float32), # joint positions of the torso
            "joint_velocity_arm_left": tf.Tensor(6, dtype=float32), # joint velocities of the left arm
            "joint_velocity_arm_right": tf.Tensor(6, dtype=float32), # joint velocities of the right arm
            "last_action": tf.Tensor(26, dtype=float32), # history of the last action
        },
        # action dimensions:
        # 26 = 6 (left arm) + 1 (left gripper) + 6 (right arm) + 1 (right gripper) + 6 (torso) + 6 (base)
        "action": tf.Tensor(26, dtype=float32),  # robot action, consists of [6x joint velocities, 1x gripper position]
        "segment_idx": tf.Scalar(dtype=int32),  # index of the segment in the episode
        "variant_idx": tf.Scalar(dtype=int32), 
    },
}

Example

We provide an example script to load our RLDS dataset and transform some episodes into mp4 video format (head camera).

import tensorflow_datasets as tfds
import tyro
import os
import imageio
from tqdm import tqdm

def main(
    dataset_name: str, 
    data_dir: str, 
    output_dir: str = "extracted_videos",
    num_trajs: int = 10
):
    ds = tfds.load(dataset_name, split='train', data_dir=data_dir)
    print(f"Successfully loaded dataset: {dataset_name}")

    os.makedirs(output_dir, exist_ok=True)
    print(f"Videos will be saved to: {output_dir}")

    for i, episode in enumerate(tqdm(ds.take(num_trajs), total=num_trajs, desc="Exporting videos")):
        head_frames = []
        
        for step in episode['steps']:
            head_rgb_image = step['observation']['image_camera_head'].numpy()
            head_frames.append(head_rgb_image)
            instruction = step['language_instruction'].numpy().decode('utf-8')

        video_path = os.path.join(output_dir, f"traj_{i}_head_rgb.mp4")
        try:
            imageio.mimsave(video_path, head_frames, fps=15)
            print(f"Saved video for episode {i} to {video_path} with instruction: '{instruction}'")
        except Exception as e:
            print(f"Error saving video for episode {i}: {e}")

if __name__ == '__main__':
    tyro.cli(main)

📜 Citation

All the data and code within this repo are under CC BY-NC-SA 4.0. If you use our dataset or models, please cite:

@article{galaxea2025,
  title={Galaxea G0: Open-World Dataset and Dual-System VLA Model},
  author={Galaxea Team},
  journal={arXiv preprint arXiv:XXXX.XXXXX},
  year={2025}
}