Human-to-human interactions can serve as a template to make the behavior of robots more natural and human. Imitation learning algorithms and the adaptation of existing motion prediction networks (Recurrent Neural Networks) could be used to develop different approaches that lead to a better prediction of human-to-human interactions, which could then be transferred to a robot in a simulation. Here, it could be demonstrated how the algorithms would work in a human-to-robot interaction. The approaches were compared with each other, and thus, the strengths and weaknesses of the different deep neural networks could be determined.