Jordan Maxwell, SCOUT Space Inc.
Keywords: Light Curve, Photometry, Space Target, Spacecraft, Machine Learning, Shape Reconstruction, Light Curve Inversion
Abstract:
An extremely efficient method for synthetically generating high-fidelity space target photometric data is presented with applications to light curve analysis including attitude inversion, spacecraft shape estimation, and object association. An enabling feature of the technology is a model for rapidly determining where, for example, a deployed solar panel will cast a shadow on the spacecraft hub based on an appropriate-fidelity shape model. Already, simulated light curves have been compared — with negligible computational time and effort — to show that major discrepancies arise when shadows are ignored even for a simple box-and-panel spacecraft. The presented technique represents a spacecraft’s geometry via a macroscopic Finite Element Method (FEM), calculates where shadows are cast given the sun angle, and determines how much total light is reflected to an observer subject to its perspective on the target. Notably, the shadowing calculation is not only efficient, but also exact subject to model fidelity (i.e. no approximations exist in calculating shadow positions beyond those made in modeling the spacecraft geometry).
Past light curve analysis approaches are extensive, but have focused primarily on 1) convex spacecraft which cast no self-shadows and therefore provide minimal information for light curve analysis 2) simplifying shape approximations (e.g. spherical geometries) with associated analysis to determine rough intelligence products, or 3) application of FEM techniques accelerated via parallelized compute architectures. Scout’s technique improves on all of these by providing a means of precisely modeling photometrics, subject to the fidelity of the target geometry and reflectance models, with nearly negligible computational effort — 5 milliseconds for a single box-and-panel shadow calculation on ARM64 processor with no explicit parallelization. The technique therefore bears significant promise in generating labeled training data for Reinforcement Learning and other Machine Learning applications toward light curve analysis. The paper provides details of relevant efforts in this direction, including lessons learned and future research objectives.
Date of Conference: September 16-19, 2025
Track: Satellite Characterization