Robot Navigation with Reinforcement Learning with ROS and Gazebo

This repository contains the files for the execution of a Reinforcement Learning algorithm, i.e. the Deep Q-Learning algorithm, for the navigation of a robot, the TurtleBot3, in a simulated environment in Gazebo, using ROS.

Architecture

The project consists of the following files:

• src/my_turtlebot3_openai_example/scripts/start_deepqlearning.py which is the main file that contains the details of the DQN architecture, all the calls to the OpenAI ROS library and the function for saving the data
• src/my_turtlebot3_openai_example/config/my_turtlebot3_openai_deepqlearn_params.yaml which contains all the parameters for the DQN architecture
• src/openai_ros/src/openai_ros/task_envs/turtlebot3/config/turtlebot3_world.yaml which contains parameters related to the TurtleBot3 enviroment
• training_results/ contains the folder related to all the experiments conducted. Each folder contains:
  • plot.png image containing the three plots relating to cumulative reward and total time, with the data not normalized
  • plot-std.png image containing the three plots relating to cumulative reward and total time, with the normalized data
  • results-date.json file containing all the parameters used in the experiment, plus the results obtained

Documentation

The project contains also RobotControl_slides.pdf which is a presentation that describe that work done in this project