ases200q2/libero_object

---
license: apache-2.0
task_categories:
- robotics
tags:
- LeRobot
- LIBERO
- manipulation
- robotics-simulation
configs:
- config_name: default
data_files: data/*.parquet
---

# LIBERO_OBJECT Dataset

This is a subset of the [lerobot/libero](https://huggingface.co/datasets/lerobot/libero) dataset containing only the **LIBERO_OBJECT** tasks.

**Note:** This dataset contains ~25% of the original lerobot/libero dataset (10 out of 40 tasks), making it:
- Faster to download (~500MB instead of ~2GB)
- Quicker to train on
- More suitable for quick experimentation

## Dataset Description

LIBERO_OBJECT is part of the [LIBERO benchmark](https://libero-project.github.io/) - a simulation benchmark for robot learning. LIBERO_OBJECT consists of 10 long-horizon manipulation tasks that require:

- Multi-step reasoning
- Object manipulation
- Scene understanding
- Tool use

### Tasks Included

This dataset contains tasks 0-9 from the original dataset:

1. Put the white mug on the left plate and put the chocolate pudding in the top drawer
2. Put the white mug on the plate and put the chocolate pudding in the top drawer
3. Put the yellow and white mug in the microwave and put the moka pot on the stove
4. Turn on the stove and put the moka pot on it
5. Put both the alphabet soup and the cream cheese in the basket
6. Put both the alphabet soup and the tomato sauce in the basket
7. Put both moka pots on the stove
8. Put both the cream cheese box and the butter in the bottom drawer of the cabinet
9. Put the black bowl in the bottom drawer of the cabinet
10. Pick up the book and place it in the back compartment of the cabinet

## Dataset Structure

- **Total Episodes:** 379
- **Total Frames:** 101,469
- **Total Tasks:** 10
- **Robot:** Franka Panda
- **Resolution:** 256x256 RGB
- **FPS:** 10

### Features

- `observation.state`: Robot joint state (8 DoF)
- `observation.images.image`: Main camera view (video)
- `observation.images.image2`: Secondary camera view (video)
- `action`: 7-DoF robot actions
- `task_index`: Task ID (0-9 for LIBERO_OBJECT)

## Usage

### Using LeRobot (Recommended for Robotics)

```python
from lerobot.common.datasets.lerobot_dataset import LeRobotDataset

# Load the dataset with video support
ds = LeRobotDataset("ases200q2/libero_object")

# Access episodes with video frames
for i in range(len(ds)):
sample = ds[i]
# sample['observation.images.image'] - numpy array of shape (H, W, C)
# sample['observation.state'] - robot state
# sample['action'] - action
```

### Using Hugging Face Datasets

```python
from datasets import load_dataset

# Load the dataset
ds = load_dataset("ases200q2/libero_object")
train_data = ds["train"]

# Access individual frames
for frame in train_data:
state = frame["observation.state"] # robot joint state
action = frame["action"] # robot action
# Note: Videos are stored externally as MP4 files
# Use LeRobotDataset for automatic video loading
```

### For Training with LeRobot

```python
from lerobot.common.datasets.lerobot_dataset import LeRobotDataset
from torch.utils.data import DataLoader

# Load dataset
dataset = LeRobotDataset("ases200q2/libero_object")

# Create dataloader
dataloader = DataLoader(dataset, batch_size=32, shuffle=True)

# Training loop
for batch in dataloader:
# batch contains observations and actions
pass
```

## Citation

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

```bibtex
@article{libero2023,
title={LIBERO: Simulation Benchmark for Long-Horizon Robot Manipulation},
author={Li, Wenhao and Mo, Kumchol and Chiu, Hung-Jui and Li, Zhenjia and Xu, Huazhe and Zhu, Yixin and Bolei, Zhou and Fei-Fei, Li and others},
journal={arXiv preprint arXiv:2310.12956},
year={2023}
}
```

## License

Apache 2.0