HOLONOMIC MOBILE ROBOT NAVIGATION DESIGN USING REINFORCEMENT LEARNING ON THE ROBOT OPERATING SYSTEM PLATFORM

• Main Author: Fauzan Ridho, Muhammad
• Format: Theses
• Language: Indonesia
• Online Access: https://digilib.itb.ac.id/gdl/view/54488
• Institution: Institut Teknologi Bandung

A mobile robot needs to be equipped with good navigation capabilities so that the robot is able to move in a space and avoid obstacles to reach one position from a certain position, especially for robots that are intended to do a job such as transporter robots and service robots. Searching for the shortest path, localization and mapping, and avoiding obstacles are the main problems in navigation topics on mobile robots. So far, many approaches have been carried out in several studies for these 3 problems, including using machine learning methods. This thesis aims to create a mobile robot navigation system using machine learning methods, Reinforcement Learning, especially using the Q-Learning algorithm. It is hoped that the mobile robot will be able to carry out the shortest route search, mapping and localization functions, and be able to avoid static and dynamic obstacles in the environment. To run a system on many hardware platforms at the same time and create a control station for monitoring robots as a human machine interface (HMI) that can connect humans and robots, this research uses the Robot Operating System (ROS) middleware platform. Mobile robot use mecanum wheels so that the robot has a more flexible holonomic movement than those using differential drive. The navigation ability of the mobile robot to a new environment was achieved after 400 episodes of training were carried out in the simulation, and it appears that the mobile robot is able to find new routes when the existing route is closed or blocked, and the mobile robot is able to avoid dynamic obstacles well