Use of Artificial Intelligence for Feature Recognition and Flightpath Planning Around Non-Cooperative Resident Space Objects

Abstract

View Video Presentation: https://doi.org/10.2514/6.2021-4123.vid This paper presents a novel method to enable small chaser spacecraft to safely approach and capture a rotating, non-cooperative resident space object in on-orbit servicing or active debris removal applications. The method is a combination of a machine vision feature recognition and localization algorithm and an artificial potential field guidance law. The machine vision approach uses the You Only Look Once V5 (YOLO-V5) object detection system to recognize, classify and localize relevant satellite components such as bodies, solar panels, antennas, and thrusters in real time. The class and location of such features is handed to the guidance algorithm, which uses a combination of virtual attractive and repulsive fields to guide the approaching chaser around collision hazards such as solar panels and towards capture targets such as satellite bodies. In combination, these algorithms identify a safe and efficient trajectory for the chaser spacecraft to capture the resident space object. The paper describes in detail the machine vision and guidance algorithms, discusses the image dataset collection, training, and testing of the machine vision algorithm, and provides performance data on the integrated system gathered in a combination of hardware-in-the-loop tests and computer simulations.