In this part of the documentation, we will describe our overall integration effort that merges together the Reinforcement Learning (RL) algorithms, and relevant ROS 2 and Gazebo packages. The framework developed is gym_gazebo2, a toolkit that combines best of both worlds: compliance with the state of the art RL algorithms and all the necessary robotics tools such as diverse ROS 2 packages for kinematics and control. To monitor the advancement of the training we utilize the already mentioned Gazebo simulator.
All the experiments based on gym-gazebo2 environments can be found in ROS2Learn, which is our collection of State of the Art algorithms applied to robotics.
We will also present some the different DRL techniques for modular robotics we have tested using our open source tools.