Deep Learning-Driven Resource Allocation for MEC-Enabled UAV Collision Avoidance System

The Internet of Unmanned Aerial Vehicles (UAVs) envisions critical services like Collision detection and Avoidance among Vehicles (CAVs). These services are typically implemented at the Multi-access Edge Computing (MEC) to enable ultra-low latency communication, ensuring real-time reactions to prevent collisions. To ensure network coverage and optimal connection of UAVs to the nearest MEC host, the CAV must be deployed across all MEC hosts. However, this may impose additional demands on these hots' available computational and memory resources. We introduce an Artificial Intelligence empowered framework to optimize virtualized resource allocation to the MEC hosts. The framework harnesses the capabilities of Deep Learning (DL) for twofold pivotal objectives: (i) Forecasting UAV Density: the framework predicts the UAV density that each MEC host must accommodate. This anticipatory insight guides resource allocation, adapting it to the anticipated demand dynamics. (ii) Precision in Virtual Resource Allotment: DL is further instrumental in calibrating the precise quantum of virtual resources requisite for the collision detection application to function optimally. This approach ensures the collision avoidance system's peak efficiency without unduly taxing MEC host computational capacities. Validation of the proposed framework entails empirical analysis. The experimental outcomes underscore the precision of the prediction model and corroborate the resource allocation framework's efficacy.