Joint-Intrinsic-Motivation

Getting started:

With python 3.8 and pip, in a virtual environment:

• Install our version multi-agent particle environment: cd into the "multiagent-particle-envs" directory and run pip install -e ..
• Install the requirements: cd into the main directory and run pip install -r requirements.txt.

Running experiments

To train the model, run the command bash /scripts/train_coop_push.sh. To run the three variants in the paper (QMIX, QMIX+LIM, QMIX+JIM), change the parameters intrinsic_reward_mode and intrinsic_reward_algo:

• QMIX: intrinsic_reward_algo="none"
• QMIX+LIM: intrinsic_reward_algo="e2snoveld" and intrinsic_reward_mode="local"
• QMIX+JIM: intrinsic_reward_algo="e2snoveld" and intrinsic_reward_mode="central"

In the scripts/ directory are multiple scripts that allow to run all the experiments shown in the paper.

Check the technical appendix for hyperparameters used in our experiments.

Citation

If you use this work, please cite cthe following paper:

@inproceedings{JIM2024,
  title={Joint Intrinsic Motivation for Coordinated Exploration in Multi-agent Deep Reinforcement Learning},
  author={Toquebiau, Maxime and Benamar, Faïz and Bredeche, Nicolas and Jun, Jae Yun},
  booktitle={Proceedings of the 23rd Conference on Autonomous Agents and Multiagent Systems},
  year={2024}
}