RA-L Submission

This repository contains

1. a multi-agent implementation of PPO and is heavily based on https://github.com/zoeyuchao/mappo, where installation steps can be found.
2. a fast implementation of HetNet
3. the implementation and running scripts for Marine environment

Note

You might have to add your own path variables in order to run the codes smoothly. Please change the path of wave information in line 187 of waves.py

Installation

To install the framework,

python setup.py develop

Run

To change the level of game, please change the variable MODE defined in line 7 and 8 in the code waves.py

To train and evaluate SchedHGNN,

cd onpolicy/scripts
sh train_marine_schedhgnn_5x5.sh
sh train_marine_schedhgnn_10x10.sh

To train and evaluate MAPPO,

cd onpolicy/scripts
sh train_marine_mappo_5x5.sh
sh train_marine_mappo_10x10.sh

To train and evaluate HetNet,

cd onpolicy/scripts
sh train_marine_hetnet_5x5.sh
sh train_marine_hetnet_10x10.sh

Installation Notes

To satisfy the minimum torch requirements do:

• CPU:

pip install torch===1.9.0+cpu torchvision -f https://download.pytorch.org/whl/torch_stable.html

• GPU:

pip install torch===1.9.0+cu102 torchvision -f https://download.pytorch.org/whl/torch_stable.html

note that in case of using GPU version, you need to also install the GPU version of the DGL by: pip install dgl-cu110. For CPU mode, just do pip install dgl.