An Off-Policy Multi-Agent Stochastic Policy Gradient Algorithm for Cooperative Continuous Control

This is the code for the paper submitted to Neural Networks. It is configured to be run in conjunction with environments from the Multi-Agent MuJoCo (MAMuJoCo).

Known dependencies

Python (3.6.13), OpenAI gym (0.19.0), tensorflow (1.12.0), numpy (1.19.5), mujoco-py (1.50.1.0).

• Download and install the MAMuJoCo by following its README.
• Different from the guidance in MAMuJoCo, we use Mujoco 1.50 instead of 2.1.

Training

1. cd into the root directory.
2. Take HalfCheetah (2x3) for example, run the following command:
   python main.py --scenario=HalfCheetah-v2 --agent-conf=2x3

Testing

We have included our well-trained policies in demos/.
Take HumanoidStandup for example, you can run the command:
python test.py --load-dir=demos/HumanoidStandup --render=True