LSY Drone Racing is a course project designed to help you develop and evaluate autonomous drone racing algorithms — both in simulation and on real Crazyflie hardware.
Whether you’re new to drones or an experienced developer, this project provides a structured and practical way to explore high-speed autonomy, control, and perception in dynamic environments.
To get started, visit our official documentation.
This project builds upon several open-source packages developed by the Learning Systems Lab (LSY) at TUM.
You can explore these related projects:
- crazyflow – A high-speed, high-fidelity drone simulator with strong sim-to-real performance.
- drone-models – A collection of accurate drone models for simulation and model-based control.
- drone-controllers – Controllers for the Crazyflie quadrotor.
Each task setup — from track design to physics configuration — is defined by a TOML file (e.g., level0.toml).
The configuration files specify progressive difficulty levels from easy (0) to hard (3):
| Evaluation Scenario | Rand. Inertial Properties | Randomized Obstacles, Gates | Random Tracks | Notes |
|---|---|---|---|---|
| Level 0 | No | No | No | Perfect knowledge |
| Level 1 | Yes | No | No | Adaptive control |
| Level 2 | Yes | Yes | No | Re-planning |
| Level 3 | Yes | Yes | Yes | Online planning |
| sim2real | Real hardware | Yes | Yes | Simulation-to-reality transfer |
Throughout the semester, teams will compete to achieve the fastest autonomous race completion times.
Competition results are hosted on Kaggle — a popular machine learning competition platform.
Note: Competition results do not directly affect your course grade.
However, they provide valuable feedback on the performance and robustness of your approach compared to others.
The competition environment always uses difficulty level 2.
If your code fails the automated tests, it is likely to encounter the same issues in our evaluation environment.
For full details, refer to the documentation.