Exponential Topology-enabled Scalable Communication in Multi-agent Reinforcement Learning

This is the implementation of our paper "Exponential Topology-enabled Scalable Communication in Multi-agent Reinforcement Learning" in ICLR 2025.

Getting Started

Create Conda Environment

Install Python environment with conda:

conda create -n ExpoComm python=3.8
conda activate ExpoComm

Install Epymarl

cd src
pip install -r requirements.txt
cd ..

Install IMP environment

pip install git+https://github.com/moratodpg/imp_marl.git

Install MAgent environment

pip install magent==0.1.14
pip install pettingzoo==1.12.0
cp env/battle_v3_view7.py PATH_TO_YOUR_PETTINGZOO_ENV/pettingzoo/magent/
cp env/adversarial_pursuit_view8_v3.py PATH_TO_YOUR_PETTINGZOO_ENV/pettingzoo/magent/

To ease the environment setup, we also provide the environmental setup we used containing detailed version information in ExpoComm_env.yaml.

Acknowledgement

The code is implement based on the following open-source projects

• EPyMARL

and the following MARL environments:

• IMP-MARL
• MAgent 2

Please refer to those repo for more documentation.

Run an experiment

python src/main.py --config=[Algorithm name] --env-config=[Env name] --exp-config=[Experiment name]

The config files are all located in src/config.

--config refers to the config files in src/config/algs. --env-config refers to the config files in src/config/envs. --exp-config refers to the config files in src/config/exp. If you want to change the configuration of a particular experiment, you can do so by modifying the yaml file here.

All results will be stored in the work_dirs folder.

For example, run ExpoComm on Adversarial Pursuit with 25 predators:

python src/main.py --config=ExpoComm_one_peer_n6 --env-config=MAgent_AdvPursuit --exp-config=ExpoComm_AdvPursuit45_s0

Citing

If you use this code in your research or find it helpful, please consider citing our paper:

@inproceedings{liexponential,
  title={Exponential Topology-enabled Scalable Communication in Multi-agent Reinforcement Learning},
  author={Li, Xinran and Wang, Xiaolu and Bai, Chenjia and Zhang, Jun},
  booktitle={The Thirteenth International Conference on Learning Representations (ICLR)},
  year={2025}
}