Autonomous visual navigation for spacecraft on-orbit operations

Abstract

Space robotic missions with increased levels of autonomy are being pursued in wide-array of orbital applications including on-orbit servicing (OOS), on-orbit assembly (OOA), and active debris removal (ADR). In these missions, the spacecraft is expected to perform most guidance and navigation tasks such as far-, mid- and close-range rendezvous, relative navigation, and proximity operations with minimal human-in-loop. This goal brings the focus towards vision-based spacecraft navigation utilising the state-of-the-art technologies in the field of computer vision especially the deep-learning algorithms for the pose estimation. This chapter explores major deep-learning approaches suitable for spacecraft pose estimation along with the discussion on different software simulation tools that are currently used for rendering realistic images of the target in orbit to train and validate the deep-learning models, and finally the ground-based testbed is used for validating the close-proximity operations.