Research on YOLOv8 Application in Bolt and Nut Detection for Robotic Arm Vision

Abstract

Robotic arms have the potential to replace human labour in various tasks, including the automated assembly of bolts and nuts. The initial step in automating this process involves detecting these objects within the robot's workspace. However, there is a noticeable scarcity of comprehensive information regarding object detection methods for bolts and nuts in the context of robotic vision. This paper aims to bridge this gap by conducting a thorough evaluation of the performance of different YOLOv8 models and dataset sizes under diverse practical conditions. We trained five YOLOv8 models and prepared 20 test datasets for evaluation. In scenarios that closely resemble the training dataset, all YOLOv8 models consistently achieved a mean average precision of 90%. Notably, YOLOv8n, YOLOv8l, and YOLOv8x delivered a remarkable overall mean average precision of over 60% across all test datasets, especially within a distance range of 20 cm to 60 cm.