Synchronisation of networks composed by fully actuated robot manipulators has received a lot of attention from the control theory community. Unfortunately, the case of under-actuated robots has not been deeply studied. The aim of this paper is to extend previous results reported by the authors addressing the particular (but of practical interest) case of networks composed by flexible-joint robots. The main feature of the contribution is to consider a change of coordinates, first introduced in Nuno, Ortega, Basanes and Hill, to solve the consensus problem assuming the existence of unknown delays in the communication channels. The extension is twofold, namely: the presentation of a control scheme that, in contrast to the one considered in Avila-Becerril and Espinosa-Perez, does not require knowledge of the initial conditions and the proof that the controller can also be implemented in Cartesian coordinates. The usefulness of the contribution is illustrated through numerical simulations.