FlightBench is an open-source comprehensive benchmark for planning methods on ego-vision-based navigation for quadrotors built on Flightmare. FlightBench provides cusomizable test scenarios (including three quantitative task difficulty metrics), representative planning algorithms, and comprehensive evaluation metrics. FlightBench also integrates the MAPPO algorithm, facilitating the training of RL-based planning methods.
For usage and more details, please refer to the documentation
[2025-03-01] 🚁 We have added real-world experiments. For Experimental results and demonstrations, please refer to our website and video.
[2025-03-01] 🏕️ We have added two more realistic environments. (See our website).
[2025-03-01] 📎 We have added a new learning-based baseline NPE (See our paper).
[2025-03-01] 💡 We have added more training details of learning-based methods to our website.
[2025-03-01] ⏩ We have added Summary Video with additional real-world experiments demos.
[2024-11-24] 🎁 We have added our implementation details and full experimental results to our website.
Please cite our paper if you use FlightBench in your work.
@article{yu2024flightbench,
title={FlightBench: Benchmarking Learning-based Methods for Ego-vision-based Quadrotors Navigation},
author={Yu, Shu-Ang and Yu, Chao and Gao, Feng and Wu, Yi and Wang, Yu},
journal={arXiv preprint arXiv:2406.05687},
year={2024}
}The source code is released under GPLv3 license.
All third party libraries we used are listed bellow:
| repository | License |
|---|---|
| flightmare | MIT |
| Fast-Planner | GPLv3 |
| ego-planner | GPLv3 |
| TGK-Planner | GPLv3 |
| sb_min_time_quadrotor_planning | GPLv3 |
| agile_autonomy | GPLv3 |