Process automation is a widespread concept and a popular approach to considering production facilities from the perspective of Industry 4.0. Using robots in a manufacturing line is one of the first options to increase efficiency, resilience, and precision, even in the conceptual envisioning stage of factory development. However, complete automation deprives the process of adaptability and robustness, a problem that can be overcome by using automation components that can perform behaviors close to human perception. The solution proposed in the present research is to adopt a deep learning system for a collaborative robot arm equipped with an adaptive gripper to obtain an intelligent system capable of handling various types of objects with different shapes in an unpredictable object-sorting scenario. An approach based on ARUCO markers is proposed to envelop the workspace where objects will be located, contributing to ease of use and generality. Conclusive experimental results, performance evaluation, and comments follow.