Install SoftGym#

Borrowed from SoftGym

Install SoftGym (Ubuntu 16.04, CUDA 9.2, Nvidia driver version 440.64)#

Dependencies#

sudo apt-get install build-essential libgl1-mesa-dev freeglut3-dev

Install#

Clone SoftGym repo

git clone https://github.com/FranBesq/softgym.git

Create conda environment Create a conda environment and activate it: conda env create -f environment.yml
Compile PyFleX: Go to the root folder of softgym and run . ./prepare_1.0.sh. After that, compile PyFleX with CMake & Pybind11 by running . ./compile_1.0.sh Please see the example test scripts and the bottom of bindings/pyflex.cpp for available APIs.

Install SoftGym (using nvidia-docker, any Ubuntu and CUDA version)#

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

Prerequisite#

Install docker-ce
Install nvidia-docker
Install Anaconda
Install Pybind11 using conda install pybind11

Running pre-built Dockerfile#

First clone SoftGym repo

git clone https://github.com/FranBesq/softgym.git

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.

Now you are in the Docker environment. Go to the softgym directory, create a conda env, set PATH and compile PyFlex

cd softgym
export PATH="PATH_TO_CONDA/bin:$PATH"
export PYFLEXROOT=${PWD}/PyFlex
export PYTHONPATH=${PYFLEXROOT}/bindings/build:$PYTHONPATH
export LD_LIBRARY_PATH=${PYFLEXROOT}/external/SDL2-2.0.4/lib/x64:$LD_LIBRARY_PATH

conda create -n softgym #you can add here packages, not needed
conda activate softgym

./compile_1.0.sh

Now that PyFleX has properly compiled. You can move outside docker (Ctrl+D), export the environment variables and start playing with the examples.

cd repos/softgym

export PATH="PATH_TO_CONDA/bin:$PATH"
export PYFLEXROOT=${PWD}/PyFlex
export PYTHONPATH=${PYFLEXROOT}/bindings/build:$PYTHONPATH
export LD_LIBRARY_PATH=${PYFLEXROOT}/external/SDL2-2.0.4/lib/x64:$LD_LIBRARY_PATH

conda activate softgym
#Running an example

python examples/random_env.py --env_name ClothFlatten

#Probably missing a lot of packages, to install them:
conda install -c conda-forge pkgname
#For example
conda install -c conda-forge numpy

If running the example fails to import softgym.* this is probably due to PYTHONPATH issues and you should make sure the interpreter knows where to look for softgym package. More info on PYTHONPATH.