Autonomous Vehicle Robot Performing Path Following by RTK-GNSS and Obstacle Avoidance by Deep Learning

Abstract

In recent years, research and development of autonomous parking systems has been conducted. The purpose of this paper is to develop an autonomous vehicle robot that can be applied to an autonomous parking system, which performs path following using RTK-GNSS (Real Time Kinematic - Global Navigation Satellite System) and obstacle avoidance using deep learning. RTK-GNSS was adopted because a highly accurate positioning system is necessary for path following. In addition, an object detection model based on YOLOv4-tiny deep neural network was employed to avoid obstacles on the path. First, the usefulness of RTK-GNSS was verified by comparing the positioning error between RTK-GNSS and stand-alone positioning. Next, the object detection model was trained to detect red traffic cones used as obstacles. Finally, an autonomous driving system for a vehicle robot was developed using RTK-GNSS and the object detection model. Then, autonomous driving experiments including obstacle avoidance were conducted to verify the usefulness of the autonomous driving system.