README.md

---
license: mit
task_categories:
- robotics
- reinforcement-learning
tags:
- drone
- simulation
- liftoff
- imitation-learning
- robot-learning
size_categories:
- n<1K
- 1K<n<10K
- 10K<n<100K
configs:
- config_name: default
  data_files:
  - split: train
    path: "chunk-*/data/*.parquet"
---

# liftoff_drone_dataset

## Dataset Description

This dataset contains telemetry and control data from the Liftoff drone simulator, recorded for robot learning and imitation learning tasks.

### Dataset Summary

- **Robot Type**: liftoff_drone_simulator
- **Total Episodes**: 1
- **Total Frames**: 1357
- **Recording FPS**: 30.0 Hz
- **Video FPS**: 30.0 Hz
- **LeRobot Version**: v3.0
- **Chunk Size**: 100 episodes per chunk

### Data Structure

The dataset is organized into chunks for efficient storage and retrieval:

```
dataset/
├── chunk-000/          # Episodes 0-99
│   ├── data/          # Parquet files with telemetry
│   ├── videos/        # MP4 videos (if available)
│   └── meta/          # Episode metadata
├── chunk-001/          # Episodes 100-199
│   └── ...
└── meta/              # Global metadata
    ├── info.json      # Dataset information
    └── stats.json     # Statistics
```

## Features

### Observation State (16 dimensions)

The observation state contains drone telemetry:

| Index | Feature | Description |
|-------|---------|-------------|
| 0-2 | pos_x, pos_y, pos_z | Position in 3D space (meters) |
| 3-6 | ori_x, ori_y, ori_z, ori_w | Orientation quaternion |
| 7-9 | vel_x, vel_y, vel_z | Linear velocity (m/s) |
| 10-12 | ang_vel_x, ang_vel_y, ang_vel_z | Angular velocity (rad/s) |
| 13-15 | acc_x, acc_y, acc_z | Linear acceleration (m/s²) |

### Action (4 dimensions)

RC controller inputs:

| Index | Feature | Description |
|-------|---------|-------------|
| 0 | throttle | Throttle input |
| 1 | yaw | Yaw control |
| 2 | pitch | Pitch control |
| 3 | roll | Roll control |

### Video Data

Each episode includes synchronized video at 30.0 FPS.
- **Format**: MP4
- **Codec**: mp4v
- **Location**: `chunk-XXX/videos/episode_XXXXXX_liftoff.mp4`

## Statistics

### Observation State Statistics

| Feature | Mean | Std | Min | Max |
|---------|------|-----|-----|-----|
| pos_x | 1108.4283 | 45.6066 | 1062.3221 | 1266.5591 |
| pos_y | -1072.1070 | 45.1135 | -1180.4570 | -1014.4241 |
| pos_z | 2.3304 | 1.5687 | 0.0495 | 8.0291 |
| ori_x | -0.1632 | 0.3821 | -0.8127 | 0.9584 |
| ori_y | 0.1264 | 0.3482 | -0.6710 | 0.7001 |
| ori_z | 0.2719 | 0.5303 | -0.8301 | 0.8648 |
| ori_w | -0.3539 | 0.4580 | -0.8651 | 0.7685 |
| vel_x | 3.1551 | 16.4758 | -31.6221 | 35.7065 |
| vel_y | -0.4859 | 20.3263 | -35.7660 | 33.6974 |
| vel_z | 0.1774 | 2.3431 | -8.8202 | 11.1492 |
| ang_vel_x | 0.2769 | 1.6505 | -8.2796 | 13.7966 |
| ang_vel_y | 0.1181 | 1.6863 | -35.1017 | 35.8174 |
| ang_vel_z | -0.1881 | 1.4257 | -7.3040 | 34.0098 |
| acc_x | 0.1231 | 34.5961 | -669.9283 | 359.1005 |
| acc_y | -0.3573 | 24.0967 | -140.9671 | 253.7777 |
| acc_z | 0.1464 | 14.7532 | -63.3092 | 382.4714 |

### Action Statistics

| Feature | Mean | Std | Min | Max |
|---------|------|-----|-----|-----|
| throttle | 0.4076 | 0.6514 | -0.9951 | 0.9843 |
| yaw | -0.0242 | 0.0779 | -0.4100 | 0.2675 |
| pitch | -0.0541 | 0.1754 | -0.9043 | 0.5213 |
| roll | 0.0623 | 0.3170 | -0.9440 | 0.8151 |

## Usage

### Loading the Dataset

```python
from datasets import load_dataset

# Load from Hugging Face Hub
dataset = load_dataset("YOUR_USERNAME/YOUR_DATASET_NAME")

# Or load locally
dataset = load_dataset("parquet", data_files="chunk-*/data/*.parquet")
```

### Using with LeRobot

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

# Load dataset
dataset = LeRobotDataset("YOUR_USERNAME/YOUR_DATASET_NAME")

print(f"Total episodes: {dataset.num_episodes}")
print(f"Total frames: {len(dataset)}")

# Iterate through data
for item in dataset:
    observation = item['observation.state']  # Shape: [16]
    action = item['action']  # Shape: [4]
    # ... use for training
```

### Accessing Videos

Videos are stored as MP4 files in chunk directories:

```python
import pandas as pd
from pathlib import Path

# Load episode data
df = pd.read_parquet("chunk-000/data/episode_000000.parquet")

# Video path is stored in metadata
# Load from: chunk-000/videos/episode_000000_liftoff.mp4
```

## Dataset Creation

### Recording Setup

- **Simulator**: Liftoff Drone Simulator
- **Bridge**: ROS2 Jazzy
- **Recorder**: Custom LeRobot-compatible recorder
- **Control**: RC controller or gamepad

### Data Collection Process

1. Start Liftoff simulator
2. Launch ROS2 bridge (`liftoff_bridge_ros2.py`)
3. Start recorder with gamepad control
4. Fly manually and record episodes
5. Data automatically organized into chunks

### Quality Control

- Episodes with crashes or poor control are discarded using emergency stop
- Each episode includes full telemetry and synchronized video
- Statistics computed across all episodes for normalization

## Licensing

This dataset is released under the MIT License.

## Citation

If you use this dataset in your research, please cite:

```bibtex
@misc{liftoff_drone_dataset,
  title={Liftoff Drone Simulator Dataset},
  author={Your Name},
  year={2025},
  publisher={Hugging Face},
  howpublished={\url{https://huggingface.co/datasets/YOUR_USERNAME/YOUR_DATASET_NAME}}
}
```

## Contact

For questions or issues, please open an issue on the dataset repository.

---

**Generated with [LeRobot Liftoff Bridge](https://github.com/YOUR_REPO)**