Physics-aware LiDAR degradation modeling and 3D reconstruction pipeline for space target on-orbit servicing
- In LEO on-orbit servicing, MLI multipath, range drift, stray light, and micro-vibrations create layering, ghosting, and edge trailing in LiDAR clouds, yet there is no interpretable degradation benchmark. Physics-level models were built for multipath, drift, trailing, false alarms, and pointing jitter, composable in simulation for repeatable comparisons.
- Reconstructing directly on heavily degraded clouds amplifies outliers, so a recoverable LiDAR-only pipeline was required. Voxel downsampling with SOR/ROR outlier rejection, IMU compensation, and multi-frame fusion greatly reduced layering and ghosting for usable surface recovery.
- On the high-fidelity platform, pipeline effectiveness was quantified with Chamfer Distance (CD), RMSE, and related metrics. Experiments show that under extreme degradation, the pipeline restores reconstruction accuracy to over 80% of the ideal baseline, with centimeter-class RMSE (0.0672 m) and completeness up to 99.37%. Digital findings were fed back into degradation-parameter settings on the on-orbit servicing simulation platform, closing the loop from simulation modeling to algorithm optimization.
Abstract
Proliferating low Earth orbit (LEO) constellations and rising space debris risks necessitate high-precision perception and 3D reconstruction for on-orbit servicing (OOS). This paper investigates physics-aware LiDAR degradation modeling and reconstruction under constraints, including extreme illumination, specular materials, micro-vibrations, and thermal deformation. We first develop a degradation model characterizing MLI-induced multipath interference, ranging drift, beam-divergence edge trailing, stray light false alarms, and platform pointing drifts. This model establishes a physics-grounded benchmark for algorithm development and evaluation within high-fidelity simulations. Second, we optimize a LiDAR-only 3D reconstruction pipeline with voxel downsampling and tandem Statistical and Radius Outlier Removal (SOR+ROR). Combined with IMU-aided motion compensation, multi-frame fusion effectively mitigates point cloud layering and ghosting. High-fidelity platform evaluations with Chamfer distance, RMSE, and completeness show robust performance under extreme lighting and complex motion. Refinement restores accuracy to over 80% of the ideal baseline, achieving centimeter-level RMSE. These results inform sensor selection and pipeline design for OOS-related simulation studies.