Implementation of Reinforcement-Learning Algorithms in Autonomous Robot Navigation

The problem of autonomous robot navigation in indoor environments must overcome various difficulties such as the dimensionality of the data, the computational cost, and the possible presence of mobile objects. This chapter addresses the implementation of an algorithm for autonomous navigation of robots in indoor environments based on machine learning. It characterizes some strategies that the literature reports and specifies a Deep Q-Network reinforcement-learning algorithm to implement on the Turtlebot robotic platform of the Gazebo simulator. Besides, a series of experiments changing the parameters of algorithm to validate the strategy shows how the robotic platform, through the exploration of the environment and the subsequent exploitation of the information, creates effective route planning.