Giskard(py)

Giskard is an open source motion planning framework for ROS, which uses constraint and optimization based task space control to generate trajectories for the whole body of mobile manipulators.

Installation instructions for Ubuntu 20.04 + Noetic

ROS Workspace

source /opt/ros/noetic/setup.bash           # source ROS
mkdir -p ~/giskardpy_ws/src                 # create directory for workspace
cd ~/giskardpy_ws                           # go to workspace directory
catkin init                                 # init workspace, you might have to pip install catkin-tools
cd src                                      # go to source directory of workspace
wstool init                                 # init rosinstall
wstool merge https://raw.githubusercontent.com/SemRoCo/giskardpy/master/rosinstall/noetic.rosinstall
                                            # update rosinstall file
wstool update                               # pull source repositories
rosdep install --ignore-src --from-paths .  # install dependencies available through apt
pip3 install -r giskardpy/requirements.txt  # install python deps
cd ..                                       # go to workspace directory
catkin build                                # build packages
source ~/giskardpy_ws/devel/setup.bash      # source new overlay

Custom Bullet Bindings

Giskard uses Adrian Röfer's bullet bindings instead of the official ones, as they are much faster for our use case. Install them like this:

mkdir -p ~/libs && cd ~/libs                # choose a place where you want to build pybullet
git clone https://github.com/SemRoCo/bullet3.git
cd bullet3                                  # be sure to be in the bullet3 folder
./build_better_pybullet.sh                  # this script will also clone and build pybind11 into libs
source ~/.bashrc                            # the script adds a python path modification to your bashrc

To test your installation do:

$ python3
Python 3.8.10 (default, Nov 14 2022, 12:59:47) 
[GCC 9.4.0] on linux
Type "help", "copyright", "credits" or "license" for more information.
>>> import betterpybullet
>>>

If it doesn't work, make sure that your $PYTHONPATH includes something like /path/to/your/bullet3/build_cmake/better_python:/path/to/your/bullet3/examples/pybullet.

Alternative QP solvers

Giskard supports multiple QP solvers and will automatically use the fasted installed solver. The default is qpalm, as it is the easiest to install.

You may want to install qpSWIFT manually. It is faster than qpalm up until ~600 constraints after which qpalm becomes faster. For reference, using the default setup, the PR2 usually requires 300-500 constraints for most goals. To install qpSWIFT, follow the instructions on their github page: https://github.com/qpSWIFT/qpSWIFT.

Giskard also supports two additional solvers, which are slower than qpalm, sorted by how fast they are for Giskard's usecase:

• Gurobi: A commercial solver, slower than qpalm, but the difference decreases with rising number of constraints:
  • sudo pip3 install gurobipy
  • You can apply for a free academic license or buy one here: https://www.gurobi.com/academia/academic-program-and-licenses/
• qpOASES: Fine up until ~100 constraints, afterwards it because extremely slow: https://github.com/SemRoCo/qpOASES/tree/noetic

Tutorials

https://github.com/SemRoCo/giskardpy/wiki

How to cite

Stelter, Simon, Georg Bartels, and Michael Beetz. "An open-source motion planning framework for mobile manipulators using constraint-based task space control with linear MPC." 2022 IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS). IEEE, 2022.