huiliu123/libero-spatial-dreamzero

LIBERO-Spatial Dataset for DreamZero

This dataset contains 500 expert demonstrations from the LIBERO-Spatial benchmark, preprocessed and ready to use directly with DreamZero for training.

Dataset Overview

Property	Value
Format	LeRobot v2.0
Robot	Franka Panda
Total Episodes	500
Total Frames	62,250
FPS	20
Image Resolution	128×128
Camera Views	agentview, eye_in_hand
Tasks	10 (LIBERO-Spatial suite)

Tasks (10 Spatial Arrangement Tasks)

Each task involves picking and placing objects in spatially-specified locations:

• Pick up the black bowl between the plate and the ramekin and place it on the plate
• Pick up the black bowl from table center and place it on the plate
• Pick up the black bowl in the top drawer of the wooden cabinet and place it on the plate
• Pick up the black bowl next to the cookie box and place it on the plate
• Pick up the black bowl next to the plate and place it on the plate
• (and 5 more spatial arrangement tasks)

Data Format

The dataset follows the LeRobot v2.0 format:

libero_spatial/
├── data/
│   └── chunk-000/
│       ├── episode_000000.parquet
│       ├── ...
│       └── episode_000499.parquet
├── videos/
│   └── chunk-000/
│       ├── observation.images.agentview/
│       │   └── episode_*.mp4
│       └── observation.images.eye_in_hand/
│           └── episode_*.mp4
└── meta/
    ├── info.json
    ├── episodes.jsonl
    ├── tasks.jsonl
    ├── stats.json
    ├── modality.json
    └── embodiment.json

Observation Space

Key	Shape	Description
observation.state	(8,)	EE position (3), EE orientation/euler (3), gripper (2)
observation.images.agentview	(128, 128, 3)	Third-person camera
observation.images.eye_in_hand	(128, 128, 3)	Wrist camera

Action Space

Key	Shape	Description
action	(7,)	Delta EE pose (6D) + gripper command (1D)

Usage with DreamZero

1. Download the dataset

from huggingface_hub import snapshot_download

snapshot_download(
    repo_id="huiliu123/libero-spatial-dreamzero",
    repo_type="dataset",
    local_dir="./data/libero_spatial"
)

2. Configure DreamZero

Use the provided config groot/vla/configs/data/dreamzero/libero_relative.yaml and update the data_dir to point to your downloaded path.

3. Start Training

bash scripts/train/libero_training.sh

Citation

If you use this dataset, please cite the original LIBERO paper:

@inproceedings{liu2023libero,
  title={LIBERO: Benchmarking Knowledge Transfer for Lifelong Robot Learning},
  author={Liu, Bo and Yu, Yifeng and Liu, Yifan and Zhang, Changhao and Yang, Yiqing and Xu, Shucheng and Garg, Animesh and Stone, Peter},
  booktitle={NeurIPS},
  year={2023}
}