The current efforts in neurorehabilitation technology research are leading to novel robotic devices capable of assisting the human motor rehabilitation process in a repeatable and durable way at low costs. A current major limitation is the inability of accurately evaluating the interaction dynamics between the patient and the robotic device. In this context, it is important to asses how the patient's neuromuscular function is modulated by the robot support. Important neuromuscular variables being modulated throughout the rehabilitation treatment include: the metabolic energy consumption, the muscle activity, the compliance in the subject's muscles and joints, and the resulting moment produced on the human joints and on the assistive device joints. It is currently not possible to accurately predict how these variables vary as the patient interacts with the assistive device. In this scenario, our current research work aims to develop a multi-level modelling approach, which can accurately predict the subject-specific neuromusculoskeletal function as well as the mechanic behavior of the robotic assistive device, and the interaction dynamics emerging from the human-machine cooperation. In this manuscript we present a first study on the development of a dynamically consistent model of a motorized ankle-foot orthosis. This will be in the future combined with physiologically accurate models of the human musculoskeletal system and used to investigate the mechanism underlying human-machine interaction strategies.