Skip to content

Latest commit

 

History

History
62 lines (49 loc) · 2.1 KB

docker.md

File metadata and controls

62 lines (49 loc) · 2.1 KB
Raw

Running SoftGym inside a docker

We provide both Dockerfile and pre-built Docker container for compiling SoftGym. Part of the docker solutions are borrowed from PyFlex

Prerequisite

Running pre-built Dockerfile

  • Pull the pre-built docker file
sudo docker pull xingyu/softgym
  • Assuming you are using conda, using the following command to run docker, which will mount the python environment and SoftGym into the docker container. Make sure you have replaced PATH_TO_SoftGym and PATH_TO_CONDA with the corresponding paths (make sure to use absolute path!).
nvidia-docker run \
  -v PATH_TO_SOFTGYM:/workspace/softgym \
  -v PATH_TO_CONDA:PATH_TO_CONDA \
  -v /tmp/.X11-unix:/tmp/.X11-unix \
  --gpus all \
  -e DISPLAY=$DISPLAY \
  -e QT_X11_NO_MITSHM=1 \
  -it xingyu/softgym:latest bash

As an example:

nvidia-docker run \
  -v ~/softgym:~/softgym \
  -v ~/software/miniconda3/:~/software/miniconda3/ \
  -v /tmp/.X11-unix:/tmp/.X11-unix \
  --gpus all \
  -e DISPLAY=$DISPLAY \
  -e QT_X11_NO_MITSHM=1 \
  -it xingyu/softgym:latest bash

This solution follows this tutorial for running GL and CUDA application inside the docker. It is important to mount the conda path inside the docker container to be exactly the same as in your home machine.

  • Now you are in the Docker environment. Go to the softgym directory and compile PyFlex
export PATH="PATH_TO_CONDA/bin:$PATH"
cd PATH_TO_SOFTGYM
. ./prepare_1.0.sh && ./compile_1.0.sh

Running with Dockerfile

We also posted the Dockerfile. To generate the pre-built file, download the Dockerfile in this directory and run

docker build -t softgym .

in the directory that contains the Dockerfile.