Abstract
Drones are being increasingly used in delivery services, often in coordination with delivery trucks through networked communication links. When a drone loses network communication with the truck in dynamic environments with uncertainties (e.g., obstacles and traffic congestion), navigation adaptation and re-routing are necessary to ensure safety. In this paper, we present a novel Drone Trajectory Planning (DTP) model based on Q-Learning, designed to adapt drone delivery missions in the presence of intermittent network connectivity. Our model leverages state representations including the drone's position, proximity to congestion zones, its general direction and truck/drone network status to adapt to uncertainties. We conduct simulations within a realistic grid environment to evaluate the performance of our DTP model. The results demonstrate DTP model's robust performance in optimizing path decisions and ensuring timely deliveries, even when communication between the truck and drone is lost due to uncertainties, achieving approximately 23% better rewards compared to the state-of-the-art A* path-finding algorithm.