Autonomous UAV Base Stations Through Deep Learning Figure 6. Coverage for 500 consecutive steps for a network with 30 MUs. relatively short time frames (e.g., days) is not highly likely. FUTURE RESEARCH DIRECTIONS In this study, we consider the positioning of a single UAV-BS. However, especially to cover large areas or to increase aggregate bandwidth offered to MUs, it is imperative to utilize multiple UAV-BSs in a coordinated fashion. Hence, placement of a plurality of UAV-BSs and integration of them into SENs through learning approaches are promising research avenues. Any UAV-BS placement approach should not be designed by assuming that the inputs are always consistent or error free. In fact, measurement and/or communication errors as well as man-made intentional disinformation occur in real-life deployments. Therefore, the training model should be robust enough to handle such issues. The proposed approach provides an effective solution to the robustness problem by averaging the locations of MUs. Yet, there are many other possible threats and errors. Nevertheless, improving the robustness and stability of a UAV-BS in the face of a wide range of threats/ errors is an interesting research challenge.