Reinforcement Learning based Sequential Controller for Mobile Robots with Obstacle Avoidance

Obstacle avoidance and path planning play substantial roles in mobile robot applications. This paper has two main parts: first, an object detection algorithm based on YOLO-v4 with custom classes and depth camera in real-time has been implemented in Robot Operating System (ROS) to resolve robot obstacle avoidance issues. Then, controlling of the robot was discussed by dividing the path into three parts with their specific PID or PD controllers. The optimum parameters of each controller are then calculated by Reinforcement Learning (RL). For ensuring the precision of the obstacle avoidance method, the accuracy of distance is evaluated by the significant result of real thorough objects. Then for evaluating the controllers, the desired-and the traveled path were compared and the positional error was calculated.