A controller that solves the problem of position synchronization of two (or more) robot systems, under a cooperative scheme, in the case when only position measurements are available, is presented. The synchronization controller consists of a feedback control law and a set of nonlinear observers. It is shown that the controller yields semi-global exponential convergence of the synchronization closed loop errors. Experimental results show, despite obvious model uncertainties, a good agreement with the predicted convergence.