Minimal OpenAI gym interface for Gazebo Robots
This repository takes inspiration from the now deprecated gym-gazebo (https://github.com/erlerobot/gym-gazebo). However, this implementation is much more lightweight, facilitating easy usage with the Turtlebot like robots.
We also include an implementation of the Dueling DQN [1] RL agent with this repository for an user to get started.
This has been tested with ROS Kinetic, Ubuntu 16.04.
For the DeepRL agent, we use PyTorch 1.0.1.
Instructions
[1] Clone the repository[2] For intializing an environment, pass the location of the launch file when creating the environment. We include an example world file in the /worlds directory.
[3] Run rllearn.py with the required dependencies to begin learning on your gazebo world. Change the reward functions in the TurtlebotGym.py file to suit your needs.
References
[1] Dueling Network Architectures for Deep Reinforcement Learning, Ziyu Wang, Tom Schaul, Matteo Hessel, Hado van Hasselt, Marc Lanctot, Nando de Freitas[2] Gym-Gazebo, https://github.com/erlerobot/gym-gazebo