Reproducing the characteristics of human movements, is a crucial issue in studying motion. In the context of this work, an explicit model of synergies which can be parametrized is used for reproducing the main features of reaching motions. This paper evaluates the possibility to extrapolate learned parameters from a captured motion to new targets and shows how learning process is a key issue to ensure the robustness of parameters. another target, some parameters displayed poor capacity to extend features of the sample motion. In this article, differences between robust and non robust set of parameters are discussed. The learning process is enhanced to ensure robustness of parameters.