In the effort to make rehabilitation robots patient-cooperative, two prerequisites have to be met: One is providing the necessary amount of guidance and safety for the patient. Just as important is transparency, i.e. minimum interaction between robot and human when it is not needed. Recently, we suggested the method of generalized elasticities, which reduce undesired interaction forces due to robot dynamics by shaping optimal conservative force fields to compensate these dynamics. We now show that these conservative force fields can not only be used to minimize undesired interaction, but that they can also support and guide the patient during therapy when needed. Thus, the patient is given maximum freedom within a safe training environment, with the aim to maximize training efficacy.