Generic Reinforcement Learning Library

Copyright 2015 Wouter Caarls

This program is free software: you can redistribute it and/or modify it under the terms of the GNU General Public License as published by the Free Software Foundation, either version 3 of the License, or (at your option) any later version.

This program is distributed in the hope that it will be useful, but WITHOUT ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License for more details.

You should have received a copy of the GNU General Public License along with this program. If not, see http://www.gnu.org/licenses/.

GRL is a C++ reinforcement learning library that aims to easily allow evaluating different algorithms through a declarative configuration interface.

sudo apt-get install git cmake g++ libeigen3-dev
git clone https://github.com/wcaarls/grl.git

For the visualization, additionally install

sudo apt-get install libgl1-mesa-dev freeglut3-dev libz-dev

For the odesim environment, additionally install

sudo apt-get install libqt4-opengl-dev libtinyxml-dev libmuparser-dev libode-dev

For the RBDL environment and MUSCOD integration, additionally install

sudo apt-get install liblua5.1-dev

For the configurator, additionally install

sudo apt-get install python-yaml python-tk

For the zeromq addon, additionally install

sudo apt-get install libprotoc-dev protobuf-compiler libsodium-dev
Then download a recent (> 4.0.0) version of zeromq and run './configure, make, make install'

mkdir build
cd build
cmake ..
make

To directly perform an experiment, run

./grld ../cfg/pendulum/sarsa_tc.yaml

To start the configurator instead, run

cd ../bin
./grlc ../cfg/pendulum/sarsa_tc.yaml

GRL comes with standard visualizations for value functions, policies, and the integrated environments (e.g. pendulum swing-up, cart-pole swing-up, compass walker)

See grl.pdf