Model building from data consists of a few steps : data collection, choice of model class, structure selection, parameter estimation and model validation. In this talk, after a brief mention of such steps, the main ideas of using auxiliary information will be discussed . In the sequel, examples taken from the field of robotics will be presented where building nonlinear models was found helpful  The data are either taken from public repositories \citedata or collected in the laboratory. The model class is multivariate nonlinear autoregressive (NAR) polynomial models. As discussed, structure selection is simplified in the present context. The main difference compared to more standard procedures is the use of auxiliary information about fixed points. This influences the stages of structure selection and of parameter estimation. The final models should be helpful to produce trajectories for robots. Laboratory tests show that the models are helpful Online.