Environmental Effects on DNN Performance During the Autonomous Refueling Task
Chris Civetta, Dillon Miller, Violet Mwaffo, Donald H. Costello
Abstract
The United States Navy has announced that it intends to dramatically increase the number of uncrewed aerial vehicles (UAVs) within a carrier air wing. Leadership would like these platforms to have the ability to aerial refuel in order to increase their range and endurance. However, to accomplish this task the UAV would be required to exhibit autonomous behavior. As of November 2023, standards and methods of compliance do not exist to allow a UAV to perform tasks without a human in or on the loop. With the support of the Office of Naval Research, the United States Naval Academy (USNA) has an active research effort devoted to developing certification evidence for a deep neural network (DNN) to be used for the autonomous aerial refueling task. Prior work at the USNA found an issue with DNN performance when the sun was in the camera's field of view. This research quantifies some of the potential environmental effects and limitations of a computer vision-based DNN being used to complete the autonomous aerial refueling task.