A Deep Reinforcement Learning-Based Path Planning Algorithm for Urban eVTOL Aircraft

As a new mode of urban air mobility, the Unmanned Aerial Vehicle (UAV) is a promising transport platform to transport high-payload in a large urban environment. However, the high-density urban buildings make the planned paths of the UAV easily fall into the local minimum traps. To adapt diversative urban environment scenarios, a deep reinforcement learning (DRL) based path planning method is proposed due to its adaptability to the environment and high-dimensional exploration. Firstly, considering energy cost, path length, and flight safety, the DRL method is used to obtain the feasible initial path in the urban environment. Then, considering kinematic constraints by using the artificial potential field (APF) method, the proposed method obtains a smooth, safe, and effective path. Thirdly, compared with the deep Q-learning method, A * method, and APF method in the randomly generated map, the proposed method shows better performance on smoothness and effectiveness.