ABSTRACT:

Vision-guided automation requires more than a camera and a robotic manipulator; it requires a reliable software pipeline that converts visual information into a structured decision for downstream robotic action. This paper presents a software-centric framework developed for object detection and decision support in a vision-guided robotic sorting application. The work focuses on the programming workflow rather than the mechanical design of the manipulator. The implemented approach combines Python-based development, computer vision, a YOLOv8 detection engine, and a dataset-driven machine-learning workflow. The software pipeline includes image acquisition, data preparation, model selection, training, validation, inference, object-label interpretation, and transfer of the detection outcome to the sorting logic. The project also considers the practical integration of the software layer with a simulated robotic environment developed using the 3DEXPERIENCE ecosystem and DELMIA. Representative objects such as an apple, banana, scissors, and mouse were used to demonstrate the recognition workflow. The study establishes a modular architecture in which the perception module can be modified independently of the robot motion and application logic. This separation improves maintainability, facilitates model replacement, and supports future integration with industrial automation systems. The contribution of the work is therefore a structured programming methodology for linking deep-learning-based object detection with robotic sorting decisions