The purpose of this repository is to help students of MVA master install DeepMimic and run tests as fast as possible. We actually lost several days attempting to make these installations. I hope it wouldn't be the case for you.
We will be using Google cloud Virtual Machines.
Note that there are two parts in DeepMimic :
- First, visualising the results
- Second, launching trainings
We will show only how to launch trainings which is the most important part of the project. One advice, start testing trainings early !! it takes minimum 24 hours for each test.
- Go to Compute Engine / VM Instances
- Create a new VM
- Choose a VM with 16vCPU
- Choose Ubuntu 16.04 LTS
- Clone the repository
git clone https://github.com/hamzakabbaj/DeepMimic_Cloud_Installation.git
- install packages
sudo apt-get update
sudo apt-get install cmake build-essential clang llvm python-dev freeglut3-dev libbullet-dev libbullet-extras-dev libglew-dev swig libopenmpi-dev python3-pip
- install virtualenv
pip3 install virtualenv
- Create a virtual environement and activate it (working with a virtual environment is necessary)
virtualenv env_deep_mimic
source env_deep_mimic/bin/activate
- install python packages
pip install pyopengl mpi4py tensorflow numpy
- install requirements
cd DeepMimic_Cloud_Installation/DeepMimic
pip install -r requirements.txt
- change Makefile and install DeepMimicCore
cd DeepMimicCore
vi Makefile
# DONT FORGET TO CHANGE EIGEN PATH !!!!!
EIGEN_DIR = /home/<Name of your machine>/DeepMimic_Cloud_Installation/eigen # <=== look for the actual path depending on the name of your machine
BULLET_INC_DIR = /usr/include/bullet
PYTHON_INC = /usr/include/python3.5m
PYTHON_LIB = /usr/lib/ -lpython3.5m
- launch DeepMimicCore installation :
make python
- CONGRATULATION !! You can now launch training
cd ../
python mpi_run.py --arg_file args/train_humanoid3d_kick_args.txt --num_workers 16