Deep Learning-based Spacecraft Relative Navigation Methods: A Survey

Jianing Song,Duarte Rondao,Nabil Aouf

DOI: https://doi.org/10.1016/j.actaastro.2021.10.025

2021-10-05

Abstract:Autonomous spacecraft relative navigation technology has been planned for and applied to many famous space missions. The development of on-board electronics systems has enabled the use of vision-based and LiDAR-based methods to achieve better performances. Meanwhile, deep learning has reached great success in different areas, especially in computer vision, which has also attracted the attention of space researchers. However, spacecraft navigation differs from ground tasks due to high reliability requirements but lack of large datasets. This survey aims to systematically investigate the current deep learning-based autonomous spacecraft relative navigation methods, focusing on concrete orbital applications such as spacecraft rendezvous and landing on small bodies or the Moon. The fundamental characteristics, primary motivations, and contributions of deep learning-based relative navigation algorithms are first summarised from three perspectives of spacecraft rendezvous, asteroid exploration, and terrain navigation. Furthermore, popular visual tracking benchmarks and their respective properties are compared and summarised. Finally, potential applications are discussed, along with expected impediments.

Robotics,Artificial Intelligence,Computer Vision and Pattern Recognition,Machine Learning

What problem does this paper attempt to address?

The problem that this paper attempts to solve is the challenges and opportunities faced in applying deep - learning techniques in spacecraft relative navigation. Specifically, the paper focuses on how to use deep - learning methods to improve the performance of spacecraft autonomous relative navigation, especially in the following aspects:

1. **Spacecraft Rendezvous and Docking**: In the rendezvous and docking missions of non - cooperative spacecraft (i.e., target spacecraft without pre - installed cooperative markers or sensors), how to estimate the relative attitude and position of the target spacecraft through deep - learning techniques to achieve precise docking.

2. **Terrain Navigation**: For descent and landing missions, how to use deep - learning techniques to identify and avoid dangerous areas, and how to perform precise terrain matching and positioning.

3. **Asteroid Exploration**: In the exploration missions on the asteroid surface or near - Earth asteroids, how to use deep - learning techniques to detect and classify the surface features of asteroids, and the precise positioning of the spacecraft relative to the asteroid.

The paper points out that although deep - learning has achieved remarkable success in fields such as computer vision, there are some special challenges in applying it to spacecraft navigation, such as high - reliability requirements, lack of large - scale labeled data sets, and difficulty in image processing under extreme lighting conditions. Therefore, the paper aims to systematically investigate the current spacecraft autonomous relative navigation methods based on deep - learning, summarize the basic characteristics, main motivations and contributions of these methods, and discuss potential applications and their expected obstacles.