부산대학교 도서관

As more and more applications in smart transportation emerge, the timely and efficient collection of data in the Internet of Vehicles (IoV) has become an important issue. Dy-namically changing traffic streams make it difficult for roadside units (RSUs) to collect vehicle data. Unmanned aerial vehicles (UAVs) with high mobility can be easily deployed anywhere to compensate for the limited communication coverage of ground-based infrastructure. Based on the above considerations, we propose an air-ground cooperation framework for vehicular data collection in real-time traffic scenarios. Specifically, we maximize the data transmission success rate (DTSR) within a given time duration by simultaneously optimizing the UAV's flight trajectory, access decision, and resource allocation. This problem is non-convex and time-continuous, and cannot be solved by conventional optimization methods. Therefore, we develop a deep reinforcement learning (DRL) approach based on the twin delayed deep deterministic policy gradient (TD3) algorithm. A real road-based traffic scenario is constructed by Simulation of Urban Mobility (SUMO) in our work. Simulation results show that the proposed method outperforms the benchmarks in terms of DTSR.