Status: developing
MADDPG implementation with Ray. Models are written in PyTorch. Tricks proposed in TD3 are added to the original method.
Install the dependencies first.
pip install -r requirements.txt
You need to install the multiagent-particle-envs.
git clone https://github.com/openai/multiagent-particle-envs.git
cd multiagent-particle-envs
pip install -e .
To use the sction space continuous and make it suitable for MADDPG, you need to modify line 29 in multiagent-particle-envs/multiagent/environment.py manually (waiting for better way...):
self.discrete_action_space = False # Continuous Action Space
You can run the code with python main.py --scenario_name simple_push.