One of the main challenges in Industry 4.0 is the supervision and coordination of heterogeneous robots at runtime, especially when they have a certain level of autonomy, as seen in Autonomous Mobile Robots. In addition, autonomous robots and their digital twins are designed by private manufacturers, and their code is often inaccessible. In this study, we present a solution to anticipate and control the behavior of robots by combining Digital Twin (DT) and Artificial Intelligence (Multiagent System and Machine Learning (ML)) models and technologies in a non-intrusive way. Using DT technologies we reproduce the action model of the robot as an autonomous agent behavior, treating the robot and its DT as black boxes. We apply existing ML techniques on the logs of its actions in various situations to predict the robot actions in the near future. We illustrate the use of our framework with a conflict resolution situation for a new AMR in a factory.