Adaptive Multi-UAV Trajectory Planning Leveraging Digital Twin Technology for Urban IIoT Applications

In this paper, flying mobile computing is considered to serve terrestrial Intelligent Internet-of-Things (IIoT) in a dynamic scenario. Existing work mainly trains the trajectory model on board, leading to UAVs' endurance reduction due to excessive energy consumption for training models. And the dynamic change and characteristics of requirements have not been considered while training trajectory. Therefore, to save energy and improve the efficiency of UAVs, we use the replication and prediction ability of DT to assist the UAV in planning the optimal trajectory, and propose an Incremental and Distributed Update (IDU) mode combined with DT to optimize its energy consumption. To cope with dynamic change of requirements, a Self-adaptive Trajectory Decision (STD) scheme is proposed, which uses the DT to plan different ranks of trajectories according to the prediction result to cope with the dynamic requirements. UAVs just need to receive this trajectory model and make a simple trajectory selection according to the real-time scenario. To plan the optimal trajectory by DT, we consider using the Dueling DQN with Prioritized Experience Replay (PER) function to train while considering the characteristics of requirements. Simulation results demonstrate the effectiveness of optimization for the DT, the STD scheme can cope with different changes in requirements and each trajectory is optimal for the corresponding scenario.