In this paper, we consider a multi-robot deployment problem involving a set of robots which must realize observation tasks at different locations and navigate through a shared network of waypoints. To solve this problem, we develop a two-level approach which alternates between (a) quickly obtaining high-level schedules based on a coarse grain CP model which approximates navigation tasks as setup times between observations, and (b) generating more accurate schedules based on a fine grain CP model which takes into account all resource usage conflicts during traversals of the shared network. The low-level layer also contains an explanation module able to generate constraints holding on high-level decision variables. These constraints (or cuts) account for interferences found in the low-level solutions and which the high-level scheduler should take into account to minimize the makespan. The proposed variants of the cut generation strategy are incomplete, the aim being to obtain good quality solutions in a short time, and they differ in the way they allow to diversify search. Experiments show the efficiency of this approach and the complementarity of the cut generation schemes proposed.