This repo provides an example of ingegrating both drake and pytorch in docker images, with Nvidia's CUDA images or vanilla Ubuntu images being the base image.
The multi-stage Jenkinsfiles specify the build process for CUDA and CPU images. To run these pipelines in Jenkins and produce CUDA images, The host machine needs to have NVidia driver and nvidia-docker2 installed.
Additional ROS packages are included in the -ros images.
Images with Python packages and C++ libtorch are published at: https://hub.docker.com/repository/docker/dexai2/drake-torch/.
Images with Python packages only and no C++ libtorch are published at: https://hub.docker.com/repository/docker/dexai2/drake-pytorch/.