Degeneracy-resilient registration and loop closure detection for a tumbling non-cooperative space target

Accurate relative pose estimation is essential for on-orbit servicing of non-cooperative space targets. However, uncontrolled tumbling and sparse, noisy LiDAR point clouds degrade conventional odometry pipelines and lead to drift accumulation. To address this issue, we propose a LiDAR odometry tailored for tumbling non-cooperative targets, integrating two newly developed core modules. First, a degeneracy-resilient registration method is proposed to mitigate sensitivity to numerical degeneracy in state-of-the-art algorithms, improving registration accuracy with negligible computational overhead. Second, leveraging the attitude-overlap correlation, an overlap-driven loop-closure detection tailored to tumbling targets that effectively suppresses accumulated registration drift is provided. The LiDAR odometry employs the degeneracy-resilient registration for continuous registration and the overlap-driven loop-closure detection for loop-closure registration. Comparative simulations with conventional methods verify that the new odometry achieves higher accuracy and efficiency. Tests on both experimental and synthetic point-cloud sequences demonstrate that it maintains attitude errors consistently below 0.06 rad under constrained computational resources, confirming its suitability for autonomous on-orbit servicing tasks.