This study explores the potential development and industrial deployment of a vision-based robotic manipulation system for cup detection and Pick-and-Place tasks using a collaborative arm robot in the beverage industry. Indeed, while most of the existing study has done on laboratory environments, this study demonstrate real production implementation gap. To do this, a dedicated domain-specific dataset was collected to fine- tune a YOLOv8 model. The YOLOv8 model was selected for its balance between accuracy and real-time performance on the edge device. To enhance robustness under varying conditions, a hybrid perception strategy combining deep learning–based detection with ROI-based depth-difference subtraction was introduced.