This is a PyTorch implementation of the paper: Learning Graph-Enhanced Commander-Executor for Multi-Agent Navigation
Project Website: https://sites.google.com/view/mage-x23
You could start training with by running sh train_gridworld.sh in directory onpolicy/scripts.
Similar to training, you could run sh render_mpe.sh in directory onpolicy/scripts to start evaluation. Remember to set up your path to the cooresponding model, correct hyperparameters and related evaluation parameters.
If you find this repository useful, please cite our paper:
@misc{yang2023learning,
title={Learning Graph-Enhanced Commander-Executor for Multi-Agent Navigation},
author={Xinyi Yang and Shiyu Huang and Yiwen Sun and Yuxiang Yang and Chao Yu and Wei-Wei Tu and Huazhong Yang and Yu Wang},
year={2023},
eprint={2302.04094},
archivePrefix={arXiv},
primaryClass={cs.RO}
}