Worldwide the number of people living with stroke-related disability is increasing. Neurorehabilitation, e.g. training therapy provided by occupational and physiotherapists helps to reduce impairment and activity limitations effectively. Yet, to cope with the increasing demands, the number of physiotherapists and occupational therapists is not sufficient. The paper proposes the hypothesis that a social humanoid robot might serve as therapeutic assistant for patients during standardized training sessions, after therapists have evaluated a patient's needs, decided on and have the patient made acquainted with an individualized training program. It is first described what kind of training tasks is intended to be supported by a social humanoid robot as training assistant. Second, digitalization of those tasks is presented. Third, the process of building user models for patients and helping persons using the repertory grid approach is discussed. Finally, opportunities for motivating interactions based on these models are mentioned.