This is the code for implementing the FCMNet algorithm presented in the paper which will appear AAMAS 2022:"FCMNet: Full Communication Memory Net for Team-Level Cooperation in Multi-Agent Systems"
Python == 3.7
pip install -r requirements.txt
First modify parameters in alg_parameters.py and env_parameters.py, then call python main.py
FCMNet.py - Defines network architecture
alg_parameters.py - Algorithm parameters
env_parameters.py - Environment parameters
env_wrapper.py - Build a multiprocessing environment and communicates with subproceses via pipes
evaluation.py - Evaluate the trained model using a greedy strategy
learner.py - Update model and record performance
main.py - Driver of program
model.py - Build tensor graph and calculate flow
policy.py - Build critic and actor
runner.py - Run multiple episodes in the multiprocessing environment
install starcraft environment - https://github.com/oxwhirl/smac
Yutong Wang
Guillaume Sartoretti