Reinforcement Learning for Mobile Robot's Environment Exploration

Mobile robots are being are being applied in various industries to perform repetitive or dangerous tasks for humans to carry out. Autonomous mobile robots are more capable than automated guided vehicles (AGV) due to their ability to be adaptable to their environment which is important for exploration of unknown environments. It is difficult to program autonomous mobile robots to adapt to various situations it may face, thus machine learning can be applied to allow a mobile robot to learn how to navigate through environments by itself. Reinforcement learning is applied in this project so that a differential drive mobile robot can learn how to navigate through its environment while avoiding collision with surrounding walls and obstacles. The reinforcement learning process is simulated by using the Robot Operating System (ROS) and its simulator Gazebo. Controlled simulation environments are created using Gazebo for the purposes of training and performance testing. Simultaneous Localization and Mapping (SLAM) will be applied to generate a map of the environment. At the end of this project, the Turtlebot3 is able to map smaller controlled environments ranging between 18m2 to 27m2 without colliding with the surrounding walls.