Vision-Based Closed-Loop Autonomous Spacecraft Rendezvous with Robust Relative Pose Estimation

Abstract

This paper presents a causally coupled, closed-loop simulator for vision based autonomous spacecraft rendezvous with a tumbling, non-cooperative spacecraft, including relative orbital mechanics, target attitude dynamics, synthetic vision, and guidance control. Using a HCW equation, verified for accuracy on the sub-nanometer range, coupled to RANSAC-EPnP pose estimation, MEKF state estimation, the architecture succeeds at relative poses recovered even with 30% contaminated outlier. Testing via Monte Carlo of LQR and MPC controllers shows that successful docking is achieved at 100% for the systems tested with up to 1.5 pixels of image noise (with success falling to 85% for 4.0 pixels as trajectories diverge statistically), where closed loop examination show that sub pixel pose error is not the limiting system component; that problem is measuremens availability, falling to 1.8% with the standard LQR (100% otherwise for this metric), with the MEKF/LQR case falling to 1.8% when the target is the only feature present and covers the view field for a fill factor in range to fill, to then fall to near 0% with the predictable fill range of 6.25m and a significant portion of data becoming dead-reckoning data.