The original version of MADDPG use environment of multiagent-particle-envs, which has no longer been updated. A maintained version of these environments is provided by PettingZoo. So this repository implement MADDPG using PyTorch and PettingZoo

training and evaluation is simple and straightforward, take simple_tag for example:

python main.py simple_tag_v2  # training
python evaluate.py simple_tag_v2 1  # evaluate result saved in folder 1

more details about arguments can be found in main.py, evaluate.py or simply run python main.py --help, python evaluate.py --help

environment name	training result	evaluation result
simple_adversary
simple_tag

• implementation of openai
• implementation of shariqiqbal2810
• maddpg-mpe-pytorch