This work deals with the leader-follower and the leaderless consensus problems in networks of multiple robot manipulators. The robots are non-identical, kinematically different (heterogeneous), and their physical parameters are uncertain. The main contribution of this work is a novel controller that solves the two consensus problems, in the task space, with the following features: it estimates the kinematic and the dynamic physical parameters; it is robust to interconnecting variable-time delays; it employs the singularity-free unit-quaternions to represent the orientation; and, using energy-like functions, the controller synthesis follows a constructive procedure. Simulations using a network with four heterogeneous manipulators illustrate the performance of the proposed controller.Peer ReviewedPostprint (author's final draft