Simulation Based Verification of Drogue Detection Algorithms for Autonomous Aerial Refueling

Abstract

Drogue detection algorithms are object detection algorithms that are used to detect the drogue basket using a camera in order to automate the docking phase of aerial refueling. Deep learning techniques, particularly Convolutional Neural Networks (CNNs) are receiving more and more interest for the drogue detection problem. The verification of such networks is a recent challenge. This paper proposes a simulation environment to generate test data, a test scenario generation approach to achieve a broad coverage, and an evaluation strategy. It not only integrates them in a testing workflow, but also demonstrates them with an example case.