Robust opponent modeling via adversarial ensemble training

Repo structure:

/core: core algorithm foradversarial ensemble training, credit to Cooperative Evolutionary Reinforcement Learning for the the evolutionary RL used as part of the training pipeline

/discreteenv: code for the game scenario modified from multi-partricle-environment

Dependencies:

python3.6

pytorch 1.0+

gym==0.9.1

Environment setup

bash setup.sh

Run

python3.6 main.py