This repo contains core sources related to Robotic AI. First users are not recommended to use this repo alone. Please have a look at example projects that use this bare code as a submodule and expose and explain one particular functionality. E.g., KOMO or rai-python.
'bare code' means that this repo contains only sources, a minimal Ubuntu-specific build system, and only minimal tests. For integration in other projects, use it as a submodule and integrate it in your own out-of-source build system.
Parts of the code have there origin at around 2004 (Edinburgh). The code grew over the years to a large repo with many projects from all lab members, but a somewhat consistent scope of code shared between projects. This repo exports a selection of the code shared between projects and contains a set of representations and methods for Robotics, ML and AI. As the functionality is diverse I don't even try to explain.
I only recently started better documentation of some parts. So far, bits and pieces of documentation are scattered:
- The Wiki page contains an introduction to KOMO. There is also an older KOMO tech report on arxiv: https://arxiv.org/abs/1407.0414
- The rai-maintenence help contains various info
- The rai-python Jupyter notebooks also help understanding the underlying code
- The test main.cpp files might help as well
git clone git@github.com:MarcToussaint/rai.git
# OR, if you don't have a github account:
git clone https://github.com/MarcToussaint/rai.git
cd rai
# The following two commands depend on the config.mk -- see below
make -j1 printUbuntuAll # for your information: what the next step will install
make -j1 installUbuntuAll # calls sudo apt-get install; you can always interrupt
make -j4
make -j4 tests bin
make runTests # compile and run the essential tests
# when interested in the python bindings:
pip3 install --user pybind11 jupyter nbconvert matplotlib
To change the dependencies edit the config.mk in the root directory:
When a flag is set =0, this forces that this package is not
used. Otherwise (when set =0 is commented), a sub-folder Makefile may
set it equal to 1 and links to this package. After this you definitely
need to recompile some components. In doubt
make cleanAll
make -j4
If you pull an update, it might help to create Makefile.dep files throught the project using
make dependAll
make -j4
The rai repo has its own old-fashioned GNU make build system, which is convenient, modular and flexible for me, and compiles each source directory into its own shared lib. But I expect serious users to replace this by their own out-of-source build system.
In rai-maintenence there is a minimalistic CMake example, showing how to compile selected sources into a single shared lib.
In rai-maintenence there are also examples for testing in docker (Ubuntu 18.04 and 16.04).
To install globally: (default path is ~/z.LOCAL)
sudo make install INSTALL_PATH=/usr/local