Predator-Prey-MARL

PyTorch Implementation of MADDPG and DDPG in a Multi-Agent Predator-Prey(MAPP) environment.

Install Python Environment

Install Python environment with conda:

conda create -n pp_env python=3.10 -y
conda activate pp_env
pip install -r requirements.txt

How to Train Agents

Execute the following command to train the agents: (MADDPG for Predators and DDPG for Prey)

wandb login
python training_MADDPG.py

Provide your wandb API key when prompted. (Get one from https://wandb.com)

Our full-trained MADDPG models are saved in the ./MADDPG_DDPG_models directory.

How to Evaluate the Performances of Agents

Execute the following command to evaluate the performances of agents:

python evaluation.py

The gifs of MADDPG models in the evaluation will be saved in the ./MADDPG_gifs directory.

Results

Environment Name	Training Results
MAPP

Training Algorithm	Evaluation Results
DDPG-DDPG
MADDPG-DDPG

References

• PettingZoo-MPE
• MADDPG
• DDPG
• MPE
• OpenAI-MPE
• OpenAI-MADDPG
• https://github.com/shariqiqbal2810/maddpg-pytorch