Docking Safety Assessment and Optimization for Autonomous Aerial Refueling: A Data-Driven Method

Docking safety is a key issue in the autonomous aerial refueling (AAR), but it has not been sufficiently studied. Because the docking system is complicated with multiple disturbances and uncertainties, docking safety is difficult to analyze theoretically. Therefore, a data-driven docking safety assessment and optimization method is proposed in this article to improve the AAR docking safety and success rate. First, a comprehensive AAR docking system is established to generate abundant realistic simulation data for the data-driven framework. Then, a deep learning method is used to extract useful information from the docking data. A safety assessment network (SAN) is proposed to predict the final docking success rate according to the current docking state. A motion prediction network (MPN) is proposed to establish the mapping relationship between the expected probe position and the docking state. Furthermore, a novel probe trajectory optimization method is proposed based on the gradient of the MPN and SAN to improve the docking success rate and improve docking safety. Finally, based on the SAN, a novel safety-oriented docking strategy that can predict failed docking attempts and retreat in advance is proposed to further improve docking safety. The effectiveness of the proposed method is demonstrated by simulations.