| Summary: | Abstract Cellular-connected unmanned aerial vehicles (UAVs), which have the potential to extend cellular services from the ground into the airspace, represent a promising technological advancement. However, the presence of communication coverage black holes among base stations and various obstacles within the aerial domain pose significant challenges to ensuring the safe operation of UAVs. This paper introduces a novel trajectory planning scheme, namely the double-map assisted UAV approach, which leverages deep reinforcement learning to address these challenges. The mission execution time, wireless connectivity, and obstacle avoidance are comprehensively modeled and analyzed in this approach, leading to the derivation of a novel joint optimization function. By utilizing an advanced technique known as dueling double deep Q network (D3QN), the objective function is optimized, while employing a mechanism of prioritized experience replay strengthens the training of effective samples. Furthermore, the connectivity and obstacle information collected by the UAV during flight are utilized to generate a map of radio and environmental data for simulating the flying process, thereby significantly reducing operational costs. The numerical results demonstrate that the proposed method effectively circumvents obstacles and areas with weak connections during flight, while also considering mission completion time.
|
|---|