OpenFlight

Less is more.

Brief Introduction • Installation • How To Use • License • Acknowledgement

Brief Introduction

1. Key components:

• Rendering Engine: Unreal Engine 4
• Dynamics: JSBSim (using Skywalker X8 model)
• Middleware: AirSim (The last version, v1.8.1)

2. Multiple UAVs supported!

• mpi4py is used for multiple processes running together.
• ZMQ is used somewhere for communication between different processes or threadings.
• Unreal Engine 5.2 can also be used, as shown in Colosseum.

3. Others

• AirSim and UE are unnecessary if real-time visualization is not required.

Installation

• Anaconda virtual environment is recommended. You can create a env like this (python 3.8 is recommended!):

conda create -n sim[your env name] python=3.8

activate it:

conda activate sim

• install JSBSim and SB3 via pip

pip install jsbsim, stable-baselines3

• install AirSim via pip (version 1.8.1)

pip install numpy pandas matplotlib simple-pid colorama
pip install msgpack-rpc-python
pip install opencv-python opencv-contrib-python
pip install airsim

• install pytorch according to https://pytorch.org/.
• clone and build Unreal Engine refer to AirSim/Colosseum
• clone and build AirSim/Colosseum (please note the version)
• configure the UE project and the AirSim Plugin.

How To Use

Two different learning methods: velocity-level (v1) and actuator-level (v2).

register(id="jsbsim-uni-uav-sample-v1", entry_point="airgym.envs:JSBSim3DUniUAVEnv",)     
register(id="jsbsim-uni-uav-sample-v2", entry_point="airgym.envs:JSBSim3DALCUniUAVEnv",)   

Run the training script using Pycharm or command line:

python navp2p_v1_train.py

When validation, run:

python navp2p_v1_test.py

or:

python navp2p_v1_test_multi.py

License

MIT

---

Acknowledgement

This open-source project has referenced and drawn inspiration from many other open-source projects. For the purpose of organizing the code structure and making it clearer, we did not directly fork these repositories, but instead, organized, improved, and added new features in a new repository. The code in the repository is not entirely written by us. We extend our respect and gratitude to the developers of the following open-source projects and release this project as open-source.

1. Quessy, Alexander, Fixedwing-Airsim

2. Gordon Rennie, gym-jsbsim

3. Miocrosoft, AirSim

4. Codex Laboratories LLC, Colosseum

5. estshorter, dwa