A Multi-Gpu Edge Server-Based Uav Trajectory Optimization for Emergencies and Disasters

This paper investigates the utilization of powerful edge computing servers to significantly reduce the computation time needed to calculate the optimal unmanned aerial vehicle (UAV) trajectory. This is especially important for UAVs in order to assist authorities and first responders reach the large number of geographically distributed users who need help during emergencies and disasters. In such hard circumstances, UAVs can help in search and rescue, monitoring, delivering supplies, and providing radio access links. The optimized UAV trajectory is computed using the Travelling Salesman Problem (TSP) approach which is NP-hard and imposes computational challenges. Hence, the TSP algorithm computations are intended to be processed in a ground multigraphics processing unit (GPU) server. In this way, both computational delays and energy consumption will be reduced for the UAV especially with the dynamic environment and large number nodes required to be visited in disaster scenarios. The proposed approach is tested via simulations considering a variable number of visited nodes, and a variable number of GPUs on the edge server. Results show a significant improvement in the computation time when multiple GPUs were used compared to the traditional central processing unit (CPU); however, using more GPUs can also lead to a slightly higher computation time compared to fewer GPUs.