commonroad-geometric (crgeo) is a Python framework that facilitates deep-learning based research projects in the autonomous driving domain, e.g. related to behavior planning and state representation learning.
At its core, it provides a standardized interface for heterogeneous graph representations of traffic scenes using the PyTorch-Geometric framework.
The package aims to serve as a flexible framework that, without putting restrictions on potential research directions, minimizes the time spent on implementing boilerplate code. Through its object-oriented design with highly flexible and extendable class interfaces, it is meant to be imported via pip install and utilized in a plug-and-play manner.
The easiest way of getting familiar with the framework is to consult the tutorial directory, which contains a multitude of simple application demos that showcase the intended usage of the package.
The installation script scripts/create-dev-environment.sh
installs the commonroad-geometric package and all its dependencies into a conda environment:
Execute the script inside the directory which you want to use for your development environment.
Note: make sure that the CUDA versions are compatible with your setup.
If you want to export the rendering frames without the animation window popping up, please use the command given below.
echo "export PYGLET_HEADLESS=..." >> ~/.bashrc
You can replace .bashrc
with .zshrc
, if you use zsh