Employing Hybrid Pointer Networks With Deep Reinforcement Learning for Drone Routing in Delivery Using Public Transportation as Carriers

Since the inception of Autonomous Aerial Vehicles (AAVs), also known as drones, and the rapid growth of the AAV industry, significant steps have been taken to design drones for diverse applications. Particularly in the last decade, the industry's progress has been so swift that novel uses for drones continue to emerge regularly. One field that has demonstrated promising potential for future applications is drone-based delivery services. However, a notable challenge in utilizing drones for deliveries is their limited operational range. To address this challenge, we propose a machine learning model that leverages public transportation vehicles as carriers to extend the range of drones, integrating Hybrid Pointer Networks (HPNs) and Deep Reinforcement Learning (DRL) to solve the complex routing problem. The primary objective is to minimize the total flying distance covered by drones while adhering to the capacity limitations and scheduled routes of public transportation buses. Notably, our model ensures that no drone is serviced if the bus roof capacity is fully utilized. By employing HPNs, we effectively manage complicated interactions between drones, buses, and delivery points, simplifying the optimization process to achieve cutting-edge results. The model was carefully optimized and fine-tuned after being trained in a simulated environment to maximize its performance. Results show that combining drone delivery services with bus service significantly lowers energy usage and improves the system's capacity to fulfill delivery obligations.