Today’s production plants are widely using automation tools to increase their productivity and improve their manufacturing process, reducing production costs and wastes. However, while fixed automation reduces cost in mass production, this is not the case in low batch size production, where the effort to re-program and test the automation in advance of being used in production is required. The connectivity of underlying subsystems with the increased use of software can, in turn, convert conventional production systems into smart cyber-physical ones, capable of demonstrating increased flexibility and adaptability to changing production demands, hence creating software-enabled industrial automation, which can be scalable and reconfigurable. This study discusses an approach for enabling an automated mixed packaging workstation supporting a different mix of products. IoT data of the entire robotic station allow the creation of a digital twin model. In turn, the connection of the digital twin model to machine learning methods allows for the automation of the entire mixed packaging process, starting from the objects’ recognition to robot control for picking and placing up to the completion of the mixed package. The proposed framework is tested in a testbed coming from the food industry and related to the mixed packaging of dairy products. The preliminary results are provided in this paper and discussed, suggesting there is potential for future investigation and applications.