One of the most consistent problem in robotics nowadays is guarantee excellent performances for any kind of manipulator, in the most possible heterogeneous conditions. The aim of this thesis is to find a versatile method to compensate the position of a manipulator in order to guarantee a pose error lower than 0.5 mm. This research work, carried out in collaboration with Comau S.p.A., shows that by using a double step compensation, based on a non-parametric and kinematic model, it is possible to untie the compensation of the positioning error of a physical manipulator from its specific model (both dynamic and kinematic).