The official repo for the CoRL 2022 paper 'Learning Control Admissibility Models with Graph Neural Networks for Multi-Agent Navigation' [project page]
The ultimate goal is to provide a benchmark and a handy tool for GNN researchers to conduct evaluations properly and fairly for multi-agent tasks.
Note: The current repo is actively under maintenance.
conda create -n pygma python=3.8
conda activate pygma
# install pytorch, modify the following line according to your environment
conda install pytorch torchvision torchaudio cudatoolkit=11.3 -c pytorch
# install torch geometric, refer to https://github.com/pyg-team/pytorch_geometric
conda install pyg -c pyg
# install pyg_multiagent
pip install pyg_multiagent