Cost-Effective Real-Time Obstacle Detection and Avoidance for AGVs using YOLOv8 and RGB-D Sensors

Abstract

Recent advances in obstacle detection and avoidance technologies have significantly enhanced robotic navigation capabilities. This study presents a real-time obstacle detection and avoidance system leveraging YOLOv8 and RGB-D sensors. The system integrates Microsoft Kinect V1 to capture RGB and depth images, employing YOLOv8 for efficient real-time object detection and classification. Depth data are utilized to calculate object distances and positions, allowing accurate navigation decisions. Implemented on the Pioneer 3DX robot, the system demonstrates high efficiency, reliability, and adaptability. With a training dataset, the model achieves exceptional performance, attaining an accuracy of 92.6% across all object classes and a [email protected] of 95%, However, the system was primarily tested in structured indoor environments, which may limit its generalization to unstructured outdoor settings. This cost-effective solution offers a practical approach to enhancing autonomous navigation and obstacle avoidance in real-world applications.