Abstract
Low-altitude economy network has been enabled to serve many civil scenarios, such as Uncrewed Aerial Vehicles (UAVs)-based parcel delivery. Nonetheless, existing UAV technologies cannot fully guarantee accurate and real-time parcel delivery. The potential Digital Twins (DT) technology has ability to enhance the performance of low-altitude UAV delivery by accurate scenario imitation and derivation. Unfortunately, existing DT technologies might not perform suitable path planning for real-time UAV delivery in highly dynamic scenarios. To address this issue, we propose a Long Short-Term DT (LST-DT)-based UAV delivery framework, by analyzing different kinds of physical data for real-time and accurate delivery. First, we implement a long-term DT based on static data including delivery destinations and latency as well as delivery environment. It can assist UAVs in exploring cooperative delivery decisions with high-efficiency resource scheduling for a high successful delivery ratio. Then, we design a short-term DT with a imitation learning mechanism to explore feasible delivery paths under the imitation of UAV status. Simulation results demonstrate that our LST-DT framework reduces system latency by an average of 45.0% compared to state-of-the-art benchmarks while improving the successful delivery ratio by an average of 19.5%.