Use of observers for the inversion of nonlinear maps: an application to the inverse kinematic problem