A search-to-control reinforcement learning (RL) based motion planning framework for quadrotor agile flight.
- RL env and utils
- ros packages
- shared libs
The code will be open-sourced once the paper is published and the refactoring work is complete.
News:
- Jan 22, 2025: 3rd edition preprint paper on arXiv.
We would like to thank the following individuals and organizations for their valuable contributions and support:
- Prof. Dong for discussions, guidance, funding and equipment support
and
- Wenxuan Gao for the support on sim2real transfer debug (PX4), RL observation discussion, and assistance with real-world experiments (May to Nov, 2024).
- Dr. Yinshuai Sun for support on drone design, real-world experiments, thrust measurement, torque coefficient measurement experiments, system identification experiments, and paper revising guidance (Apr 2024 to Jan 2025).
- Dr. Yunlong Song for the discussions and support on sim2real transfer and RL environment design (May to Aug, 2024).
- Baiyang Li for support on PX4-Autopilot discussion and co-debug in rate controller, mixer and control allocator (Nov and Dec, 2024).
- Zeshuai Chen for discussions on UAV configuration, motion capture communication setup, IMU integration, as well as plotting and video creation.
- Dr. Tao Cui for assistance with thrust measurement experiments and early discussions on RL reward functions (Mar to May, 2024).
- Hongzheng Zhu for assistance with motor modeling and real-world experiments (Sept and Dec, 2024).
- Yubo Dong for discussions on RL observation space, SB3 preserving steps and hyperparameter-tuning trails (Mar to May, 2024).
- Jia Xu for discussions on UAV configuration.
- Yufan Zhou for assistance with real-world experiments (Dec, 2024).
- Jingyi Tu for support with torque coefficient measurement experiments (Oct, 2024).
- T-Motor for providing the T-Motor F60Pro Kv2550 motor parameters.
- Nokov for supporting motion capture equipment.