Readme

This package contains Gazebo plugins intended to simulate reinforcement learning algorithms in mobile robots.

Installation

Supported system: Ubuntu 16.04 Xenial
Basic dependencies: cmake, gcc, opencv

1. Install Gazebo 7

   sudo apt-get install libgazebo7-dev

2. Compile and install caffe on the system

   cmake .. -DCMAKE_INSTALL_PREFIX=/usr/local

3. Compile and install the source code

   mkdir build && cd build
   cmake .. -DCOMPILE_DRL_PLUGIN=1
   make -j4 && sudo make install

Testing

1. Edit the gazebo world file in gazebo/worlds/
   • set the mode tag to test
   • select number of iterations in max_steps
2. Launch gazebo

   gazebo gazebo/worlds/rover.world --verbose
   gazebo gazebo/worlds/nfq_rover.world --verbose
   gazebo gazebo/worlds/drl_rover.world --verbose

Training

1. Edit the gazebo world files (rover.world, nfq_rover.world, drl_rover.world):

   • set the mode tag to train
   • select number of iterations in max_steps

2. Launch gazebo rover world:

       gazebo gazebo/worlds/rover.world --verbose

Acknowledgement

This work is part of Thomio Watanabe PhD project funded by grant: #2015/26293-0, São Paulo Research Foundation (FAPESP). "Opinions, hypothesis and conclusions or recommendations expressed herein are the author(s) responsibility and do not necessarily conform with FAPESP vision."

Copyright © 2017 Thomio Watanabe