In this note we propose a simple solution to the regulation problem of rigid robots based on the availability of only joint position measurements. The controller consists of two parts: (1) a gravitation compensation, (2) a linear dynamic first-order compensator. The gravitation compensation part can be chosen to be a function of either the actual joint position or the desired joint position. Both possibilities are aproved to yield global asymptotic stability. Performance issues of the controller are illustrated in a simulation study of a two degrees-of-freedom robot manipulator.