Development of Automated Guided Vehicles Using Software Engineering

Abstract

Effective Automated Guided Vehicle (AGV) management systems are critical to reducing collision risks and ensuring reliable operation. Implementing AGV management based on software engineering (SE) principles enables robust, scalable, and secure control software. This paper presents a real-time object detection and obstacle avoidance system for a Forklift AGV, engineered using SE principles. The system integrates a camera and sensors for autonomous navigation, focusing on compatibility, maintainability, and ease of use. The development follows three key stages: long-range object detection under varying lighting conditions, precise object position estimation, and rigorous performance evaluation. The system employs YOLOXI, an enhanced version of YOLOv3, using machine learning and computer vision for real-time object detection and distance calculation to prevent collisions. Based on SE metrics and experimental testing, performance validation emphasized availability and security. The results demonstrated high detection accuracy across different weather conditions, with mean average precision (mAP) scores of 94.2% (sunny), 92.5% (rainy), and 90.3% (cloudy), enabling effective detection and collision avoidance up to 10 seconds before impact. The AGV's detection model was trained on three datasets: Pascal VOC 2012, Pascal VOC 2007, and Pascal VOC Integration (a combination of both). The model achieved mAP scores of 85.9%, 89.8%, and 94.0%, respectively, outperforming existing detection models suitable for AGV integration.