OmniDrones is an open-source platform designed for reinforcement learning research on multi-rotor drone systems. Built on Nvidia Isaac Sim, OmniDrones features highly efficient and flexible simulation that can be adopted for various research purposes. We also provide a suite of benchmark tasks and algorithm baselines to provide preliminary results for subsequent works.
For usage and more details, please refer to the documentation.
The initial release of OmniDrones is developed based on Isaac Sim 2022.2.0. As the next version of Isaac Sim (2023.1.0) is expected to bring substantial changes but is not yet available, the APIs and usage of OmniDrones are subject to change. We will try our best to keep the documentation up-to-date.
The new release of Isaac Sim (2023.1.0) has brought substantial changes as well as new possibilities, among
which the most important is new sensors. We are actively working on it at the devel branch. The release
branch will still be maintained for compatibility. Feel free to raise issues if you encounter any problems
or have ideas to discuss.
Please cite this paper if you use OmniDrones in your work:
@misc{xu2023omnidrones,
title={OmniDrones: An Efficient and Flexible Platform for Reinforcement Learning in Drone Control},
author={Botian Xu and Feng Gao and Chao Yu and Ruize Zhang and Yi Wu and Yu Wang},
year={2023},
eprint={2309.12825},
archivePrefix={arXiv},
primaryClass={cs.RO}
}
Some of the abstractions and implementation was heavily inspired by Isaac Orbit.