Development of a Robotic Cell Using a YOLOv8 Vision System Embedded in a Raspberry Pi

Abstract

Artificial intelligence has driven significant advancements in industrial applications, particularly in automation and process control. Deep learning models, such as convolutional neural networks (CNNs), have been particularly effective in these applications. This study presents a Pick and Place solution utilizing a SCARA robotic manipulator and a computer vision system based on the YOLOv8 model. Communication between the computer vision system and the SCARA robot is achieved via socket TCP, enabling real-time detection of parts and optimizing the precision of robotic operations. The five YOLOv8 models trained were evaluated, with the highest accuracy model achieving 96.7%. Using the Raspberry Pi 4 to embed the vision system, the inference time for all models was greater than 800 ms. Thus, to use it in an industrial scenario, the inference time should be analyzed.