Optimal Computer Vision-Based Spacecraft Rendezvous and Proximity Operations in the Cislunar Region

The capability to perform rendezvous and proximity operations (RPO) is critical for a wide range of future cislunar missions. These operations underpin key applications in the emerging in-space economy, including orbit-transfer vehicles (OTVs), active debris removal (ADR), and on-orbit servicing, assembly, and manufacturing (OSAM). Stellerian has developed a neural-network-based vision sensor capable of estimating the relative pose of uncooperative targets. This work integrates these measurements into a unified estimation–control pipeline. An unscented Kalman filter (UKF) is used to recover the full relative state, and the estimated state is subsequently used within a model predictive control (MPC) framework to generate optimal, constraint-satisfying trajectories. Simulation results demonstrate that the vision-enabled UKF achieves position errors below 0.1 m and velocity errors under 0.01 m/s, along with highly accurate attitude estimates. These results confirm the feasibility of combining uncooperative pose estimation with optimal control to enable autonomous and precise RPO in the complex cislunar environment.