python bindings to rai

This repo builds python bindings to rai, including a Pypi wheel.

Documentation

Although very incomplete, the best intro to the code is found as part of the robotics-course documentation. Jupyter notebooks that demonstrate the use are found in the rai tests and in tutorials/.

Installation via pip

• The pip package was compiled for python3.6 .. 3.10, and most of the dependencies statically linked. A few are still loaded dynamically, which requires installing on Ubuntu:

sudo apt install liblapack3 freeglut3 libglew-dev python3 python3-pip

• pip-install robotic and dependencies (numpy, scipy)

python3 -m pip install --user robotic numpy scipy

• Test:

python3 -c 'from robotic import ry; print("ry version:", ry.__version__, ry.compiled());'
python3 -c 'from robotic import ry; ry.test.RndScene()'

If the rai-robotModels path fails, find rai-robotModels and try something like

python3 -c 'from robotic import ry; ry.setRaiPath("/usr/local/rai-robotModels"); ry.test.RndScene()'

When rai-robotModels is still messed up, try cloning it completely:

cd ~/.local; rm -Rf rai-robotModels;
git clone https://github.com/MarcToussaint/rai-robotModels.git

tested within a ubuntu:latest docker:

sudo apt install --yes liblapack3 freeglut3 libglew-dev python3 python3-pip
python3 -m pip install --user robotic numpy scipy
python3 -c 'from robotic import ry; ry.test.RndScene()'

Installation from source

This assumes a standard Ubuntu 20.04 (or 18.04) machine.

• Clone the repo:

git clone --recursive https://github.com/MarcToussaint/rai-python.git
cd rai-python

• Install Ubuntu packages. The following should do this automatically; if you don't like this, call make -j1 printUbuntuAll to see which code components depend on which Ubuntu packages, and install by hand.

sudo apt-get update
make -C rai -j1 installUbuntuAll  # calls sudo apt-get install; you can always interrupt

• Install python packages, including pybind:

echo 'export PATH="${PATH}:$HOME/.local/bin"' >> ~/.bashrc   #add this to your .bashrc, if not done already
sudo apt-get install python3 python3-dev python3-numpy python3-pip python3-distutils
pip3 install --user --upgrade pip
pip3 install --user jupyter nbconvert matplotlib pybind11

(If tab-autocomplete for jupyter does not work, try pip3 install --user jedi==0.17.2 )

• Compile using cmake:

ln -s build_utils/CMakeLists-ubuntu.txt CMakeLists.txt
make -C rai cleanAll
make -C rai unityAll
mkdir build
cd build
cmake ..
make -j $(command nproc)

• Test a first notebook, then checkout all notebooks in notebooks/ and rai/test/ry

jupyter-notebook tutorials/1-basics.ipynb

• Other tests

cd examples
python3 skeleton-solving-example.py

Building a wheel within a manylinux docker

• Build the docker

cd build_utils
./build-docker.sh

• Run docker and compile wheels inside

./run-docker.sh
## inside docker:
cd local #this mounts rai-python/
make -C rai -j1 unityAll
build_utils/build-wheels.sh
exit

• Outside of docker, install locally with pip or push wheels to pypi

# e.g.
python3.8 -m pip install --user dist/robotic-*cp38*.whl --force-reinstall
python3.10 -m pip install --user dist/robotic-*cp310*.whl --force-reinstall
# or
twine upload dist/*.whl

Use of the wheel binary in C++

• Get the binary lib by installing the pip package:

python3 -m pip install --user robotic

• Get the sources by cloning this repo recursively:

cd $HOME/git; git clone --recursive https://github.com/MarcToussaint/rai-python.git

• Copy things into an include and link folder (like 'make install') CHANGE PYTHON VERSION:

mkdir -p $HOME/opt/include/rai $HOME/opt/lib
cp $HOME/.local/lib/python3.6/site-packages/robotic/libry.so -f $HOME/opt/lib/libry.cpython-36m-x86_64-linux-gnu.so
cp $HOME/git/rai-python/rai/rai/* -Rf $HOME/opt/include/rai
cp $HOME/git/rai-python/botop/src/* -Rf $HOME/opt/include/rai

• Compile your main

gcc script2-IK.cpp -I$HOME/opt/include/rai -L$HOME/opt/lib -lry.cpython-36m-x86_64-linux-gnu -lstdc++ `python3-config --ldflags`