Low-noise trajectory optimization of urban air mobility in the urban environment using deep reinforcement learninga)

This study proposes a method for optimizing low-noise flight trajectory for urban air mobility (UAM) in urban environments using deep reinforcement learning (DRL). The objective is to efficiently derive flight trajectories that minimize noise impact on ground observers as UAM vehicles approach a vertiport for landing. Noise data were calculated for various flight velocities, and a deep learning (DL) model was employed to estimate noise map from noise propagation. A DRL model based on the Soft actor critic algorithm was designed, incorporating noise reward function through the trained DL model while adhering to specified flight constraints. The results confirmed that the DRL model effectively guided the UAM vehicle to the target area while accomplishing the given flight conditions. Various case studies were conducted and confirmed that the DRL model can optimize unique noise-optimized trajectories for a given simplified urban environment, vehicle initial state, and designated noise reduction zone. Also, the optimized low-noise trajectory was shown to reduce both the average noise level and the proportion of high noise impact areas compared to the reference trajectory. This DRL based trajectory optimization method is expected to contribute to the development of low-noise operational strategies for UAM vehicles in urban settings.