Grasping Learning Environment

Grasping learning environment simulates grasping tasks in Gazebo. The environment is written as a Python module with the same interface as the environments used in OpenAI's Gym.

How to cite

For details check out the paper "Leveraging Contact Forces For Learning to Grasp" on arXiv.

@inproceedings{2019_ICRA_mbkrb,
  author    = {Hamza Merzic and
               Miroslav Bogdanovic and
               Daniel Kappler and
               Ludovic Righetti and
               Jeannette Bohg},
  title     = {Leveraging Contact Forces for Learning to Grasp},
  booktitle = {2019 IEEE International Conference on Robotics and Automation},
  year      = {2019},
  url       = {http://arxiv.org/abs/1809.07004}
}

License

• "baseline" (Copyright (c) 2017 OpenAi) in src/baselines_catkin/baselines licensed using The MIT License, see related folder for licensing text
• "catkin" (Copyright (c) 2012 Willow Garage, Inc) in src/catkin/ licensed using the Software License Agreement (BSD License). See related folder for licensing text
• "Boost python" (Copyright © 2002-2015 David Abrahams, Stefan Seefeld) in src/gazebo_grasping/external/lib/ licensed using the Boost Software License - Version 1.0. See related folder for licensing text
• Other content (Copyright (c) 2018 Max Planck Gesellschaft) licensed using the BSD 3-Clause Licence

Installation

In order to install the GLE follow the steps below:

# Clone the repo.
git clone git@git-amd.tuebingen.mpg.de:amd-clmc/grasping_sim.git

# Install Gazebo8 and system dependencies.
sudo sh -c 'echo "deb http://packages.osrfoundation.org/gazebo/ubuntu-stable `lsb_release -cs` main" > /etc/apt/sources.list.d/gazebo-stable.list'
wget http://packages.osrfoundation.org/gazebo.key -O - | sudo apt-key add -
sudo apt-get update
sudo apt-get install -y gazebo8 libgazebo8-dev cmake protobuf-compiler python3-pip

# Install Python dependencies.
sudo pip3 install gym catkin_pkg catkin_tools trollius empy mpi4py tensorflow scipy

Next step is to create your grasping workspace, link the src folder to the workspace and build the module, e.g.

export GRASPING_WS=$HOME/grasping_ws
mkdir $GRASPING_WS
cd $GRASPING_WS
ln -s /path/to/grasping_sim/src src

catkin init
catkin build -DCMAKE_BUILD_TYPE=Release -DPYTHON_VERSION=3.5 -DPYTHON_EXECUTABLE:FILEPATH=`which python3`

# Set up environment variables.
echo "export GRASPING_WS=$GRASPING_WS" >> $HOME/.bashrc
echo "export GAZEBO_MODEL_PATH=\$GRASPING_WS/src/gazebo_grasping/models" >> $HOME/.bashrc
echo "export GAZEBO_PLUGIN_PATH=\$GRASPING_WS/build" >> $HOME/.bashrc
echo "source \$GRASPING_WS/devel/setup.bash" >> $HOME/.bashrc
echo "source /usr/share/gazebo/setup.sh"
source $HOME/.bashrc

Test example

First test if Gazebo is properly installed by running:

gazebo --version

This should write out Gazebo multi-robot simulator, version 8.1.1.

Then open a terminal and run python3. The interpreter should start and input the following:

from grasping_env import GraspingEnvPregrasps
env = GraspingEnvPregrasps()
env.render()

At this point you should have a few information messages printed out. Ignore the warning that the objects will float - this is the intended behavior since the gravity is turned off.

Now, open another terminal and run gzclient. You should see a robot hand along with an object in the middle. You can now control the simulation from the python. For example, to move the fingers you can do:

env.step([1, 1, 1, 1])
env.step([1, 1, 1, 1])
env.step([1, 1, 1, 1])
env.step([1, 1, 1, 1])

You can restart the environment by doing:

env.reset()

Before exiting don't forget to close the environment by doing:

env.close()