Simulation-based performance assessment of ORB, YOLOv8, and picking strategies for single-arm robot conveyor belt pick-and-place operations

Pick-and-place robots play a crucial role in industrial automation, helping to lower labor costs, minimize errors, and improve production efficiency. Many image processing methods have been proposed to facilitate the pick-and-place operation. However, the performance of these methods is sensitive to the lighting conditions, presence of occlusions, and variations in the object appearance. Although many of these challenges can be overcome through the use of deep learning methods, a direct performance comparison of image processing methods and deep learning methods, coupled with an analysis of different picking strategies, is lacking. The present study addresses this gap by conducting a simulation-based evaluation of the accuracy and processing time of the ORB image processing algorithm and YOLOv8 deep learning model for object recognition. The effects of two different picking strategies (FIFO and Euclidean Distance) on the system throughput are also explored. The simulation results show that YOLOv8 achieves a higher accuracy (98%) and significantly faster processing time (138 ms) than ORB (97.33% accuracy and 715.24 ms processing time). Additionally, the FIFO picking strategy improves productivity by 13% compared with the Euclidean Distance strategy. Overall, the findings provide valuable insights into optimizing robotic pick-and-place operations in industrial automation settings.