The energy-shaping methodology introduced by Takegaki and Arimoto in 1981 is a powerful tool for global regulator design for rigid robot manipulators. In this paper we extend this approach to the global regulator design for elastic joint robots. Under this framework we broach the cases when rotor positions and velocities are available for measurement as well as when only rotor positions can be measured. The structure of these regulators is given by the gradient of a suitable class of artificial potential energies and damping injection via rotor velocity or an appropriate filtering of rotor position.