In this paper, we consider the problem of consensus of a network of heterogeneous nonlinear agents on a family of different desired trajectories generated by an uncertain leader. We design a set of local reference generators and local controllers, which guarantees that the agents achieve consensus robustly on all possible trajectories inside this family. The design of the local reference generators is based on the possibility to express the trajectory of the leader as a nonlinear regression law, which is parametrized by some constant unknown parameters.