We address the problem of devising an optimized energy aware flight plan for multiple Unmanned Aerial Vehicles (UAVs) mounted Base Stations (BS) within heterogeneous networks. The chosen approach makes use of Q-learning algorithms, through the definition of a reward related to relevant quality and battery consumption metrics, providing also service overlapping avoidance between UAVs, that is two or more UAVs serving the same cluster area. Numerical simulations and different training show the effectiveness of the devised flight paths in improving the general quality of the heterogeneous network users.