Now, the experiment video can be accessed by downloading them from ExperimentsVideo.zip
Since we use LFS to upload videos, if you want to clone project including videos, use git lfs clone to pull this repo instead of git clone.
If you only want to access the code, use git clone, and it won't pull videos from LFS.
Source code of algorithm can be accessed.
By the compiled MADDPG.exe, you can train your own model !!!
import this repo to omnet++, choose MADDPG.exe, then [run as] --> [run configuration].
Choose old.ini as the ini file, you can config by revising old.ini file. Don't forget to choose cmdEnv when training model
Sadly, it's unconvienient that you should also run one algorithm with python after you launched a omnet++ simulation in it's ide.
Due to some error of omnetpp, I can not run this MADDPG.exe in python with multi-thread support.
All the communication between these two processes is implemented by .json file.
Actually, the time consuming is mainly from the simulation, and I don't optimize it yet.
The requirement is gym == 0.10.5, tensorflow == 1.8.0, matplotlib and python == 3.6
The MADDPG algorithm is revised from OpenAI https://github.com/openai/maddpg. Visit their website for more details.
Any question can contact me at: quanyili0057@gmail.com