A Fully-annotated, Open-design Dataset of Autonomous and Piloted High-speed Flight

This repository contains a dataset characterized by:

• fast (>21m/s) and aggressive quadrotor flight
• autonomous and human-piloted flight, on multiple trajectories
• high-resolution, high-frequency collection of visual, inertial, and motion capture data
• it includes drone racing gates—with bounding boxes and individual corner labels
• it includes control inputs and battery voltages

If you use this repo, you can cite the companion paper as:

@ARTICLE{10452776,
         author={Bosello, Michael and Aguiari, Davide and Keuter, Yvo and Pallotta, Enrico and Kiade, Sara and Caminati, Gyordan and Pinzarrone, Flavio and Halepota, Junaid and Panerati, Jacopo and Pau, Giovanni},
         journal={IEEE Robotics and Automation Letters}, 
         title={Race Against the Machine: A Fully-Annotated, Open-Design Dataset of Autonomous and Piloted High-Speed Flight}, 
         year={2024},
         volume={9},
         number={4},
         pages={3799-3806},
         doi={10.1109/LRA.2024.3371288}
}

Tested on October 2023 with Ubuntu 20.04 LTS (Python 3.8), macOS 14 (Python 3.11), Windows 11 (Python 3.9).

Clone the repository and install the requirements:

git clone https://github.com/Drone-Racing/drone-racing-dataset.git
cd drone-racing-dataset
pip3 install -r requirements.txt

From folder drone-racing-dataset/, use the following scripts to download the dataset files.

On Ubuntu and macOS:

sudo apt install wget   # or, e.g, `brew install wget` on macOS
sudo chmod +x data_downloader.sh
./data_downloader.sh

On Windows, double-click on file drone-racing-dataset/data_downloader.cmd

This will create and populate 2 folders in the root of the repository:

• data/piloted/
• data/autonomous/

data/piloted/ and data/autonomous/ contain 12 and 18 flight-.../ folders respectively.

For each flight, 2 CSV files are provided:

• One sampled/interpolated at the timestamps of the camera frames (..._cam_ts_sync.csv)
• One sampled/interpolated at 500Hz (..._500hz_freq_sync.csv)

Each CSV file contains the following columns:

For each flight, 2 folders contain the Arducam video capture data:

• camera_flight-.../ contains all the captured frame in JPEG format
• labels_flight-.../ contains the racing gates' bounding boxes and corner labels in the TXT format described below

Each TXT file contains lines in the form 0 cx cy w h tlx tly tlv trx try trv brx bry brv blx bly blv where:

• 0 is the class label for a gate (the only class in our dataset)
• cx, cy , w, h ∈ [0, 1] are a gate's bounding box center’s coordinates, width, and height, respectively
• tlx, tly ∈ [0, 1], tlv ∈ [0; 2] are the coordinates and visibility (0 outside the image boundaries; 2 inside the image boundaries) of the top-left internal corner. Similarly for tr, bl, br, the top-right, bottom-left, and bottom-right corners.

All values are in pixel coordinates normalized with respect to image size. The keypoints label format follows the COCO definition. The gates/lines in each TXT file are not ordered.

The scripts in the scripts/ folder can be used to visualize the data and to convert the data to other formats.

To plot one of the CSVs, for example, use:

cd scripts/
python3 ./data_plotting.py --csv-file ../data/autonomous/flight-01a-ellipse/flight-01a-ellipse_cam_ts_sync.csv

To visualize the Arducam frames and labels, use (press SPACE to advance, CTRL+C to exit), for example:

cd scripts/
python3 ./label_visualization.py --flight flight-01a-ellipse

Folder quadrotor/ contains the bill of material and STL files of the COTS/open design of the racing drone used to collect the dataset. In the same folder, you can also find the Betaflight parameters backup used to configure the drone. A tutorial to assemble the drone is on YouTube.

Tested on October 2023 with Ubuntu 20.04 LTS

ROS2 .sqlite3 bags are stored in the ros2bag_.../ folder of each flight.

The rosbags contain topic with custom messages defined in the repository drone-racing-msgs.

To play a rosbag:

• Install ROS2 Humble following the official guide
• Install the custom messages defined in drone-racing-msgs

mkdir -p ~/drone_racing_ws/src
cd ~/drone_racing_ws/src
git clone https://github.com/tii-racing/drone-racing-msgs.git
cd ..
rosdep install --from-paths src --ignore-src -r -y
colcon build --symlink-install

• Source the ROS2 workspace and play the rosbag

source ~/drone_racing_ws/install/setup.bash
ros2 bag play data/autonomous/flight-01a-ellipse/ros2bag_flight-01a-ellipse

• Print the topic list and echo a topic

source ~/drone_racing_ws/install/setup.bash
ros2 topic list
ros2 topic echo sensors/imu

drone-racing-dataset
├── camera_calibration
│   ├── calibration_results_trackRATM.json - Camera parameters of trackRATM flights in JSON format.
│   ├── calibration_results_trackRATM.npz - Camera parameters of trackRATM flights in NumPy format.
│   ├── calibration_results.json - Camera parameters of ellipse and lemniscate flights in JSON format.
│   ├── calibration_results.npz - Camera parameters of ellipse and lemniscate flights in NumPy format.
│   └── drone_to_camera.json - Drone to camera in JSON format. Translation was measured by MoCap, rotation is an estimation, and includes the FLU to RDF conversion.
├── ...
└── scripts
    ├── trajectory_generation
    │   ├── ellipse.py - Script used to generate the trajectories of autonomous ellipse flights.
    │   ├── lemniscate.py - Script used to generate the trajectories of autonomous lemniscate flights.
    │   ├── trajectory.py - Base class for trajectory generation.
    ├── camera_calibration.py - Script used to generate the files in `camera_calibration/`.
    ├── create_std_bag.py - Script used to generate standard ROS2 bags with `Image`, `Imu`, and `PoseStamped` messages. 
    │                       Standard bags are not provided in the dataset because of their size (>10GB each).
    │                       You will need the ROS2 workspace with the custom messages installed (see section "ROS2 Bags").
    ├── data_interpolation.py - Script used to generate the comprehensive CSV files interpolated at arbitrary frequencies.
    ├── ...
    ├── measure_dtw.py - Script used to measure the DTW distance between trajectories of same mode and shape.
    └── reference_controller.py - Python implementation of the PID controller used for the autonomous flights.

The dataset released in Race Against the Machine A Fully-annotated, Open-design Dataset of Autonomous and Piloted High-speed Flight by TII Racing Jetsetters is licensed under CC BY 4.0