High-Fidelity Orbital Reconstruction: A Physics-Grounded Simulation and Two-Phase Progressive Strategy for Space Objects
- On-orbit reconstruction of non-cooperative targets lacks ground-truth calibration and is severely affected by thermal drift and solar saturation. Led construction of a high-fidelity simulation platform integrating a Gemini 435Le depth camera, Robosense E1R solid-state LiDAR, and IMU, with refined physical sensor error models to reproduce typical on-orbit degradations and provide a unified experimental benchmark for strategy comparison.
- Single-pass scans struggle to balance completeness and precision. A “manifold-to-point” two-stage pipeline was implemented: stage one uses spherical scanning for 100% topological coverage; stage two applies active close-range refinement, improving Chamfer Distance by 7.4%. Dual statistical denoising suppresses orbital artifacts, achieving local reconstruction accuracy of 1.15 cm and supporting measurable OOS task outcomes.
- The pipeline had to be interpretable and measurable for proximity and robotic tasks in congested orbits. Metric design and benchmarking against baselines yielded a reproducible framework and quantitative evidence backing the paper’s on-orbit servicing perception simulation claims.
Abstract
Safe On-Orbit Servicing (OOS) demands high-precision 3D reconstruction of non-cooperative targets. We develop a simulation platform incorporating detailed error models for LiDAR, depth cameras, and IMUs, accounting for thermal drift, solar saturation, and multi-path interference. A proposed two-phase “manifold-to-point” strategy integrates global spherical scanning for topological completeness with an active approach for local refinement. After suppressing orbital artifacts via edge-preserving statistical denoising, experiments demonstrate nearly 100% topological coverage and centimeter-level accuracy. This framework facilitates autonomous proximity operations and robotic manipulation in congested orbits.