Characterization Of Neural Ordinary Differential Equations For Astrodynamics Applications
Abstract
This work investigates the potential of neural ordinary differential equations (neural ODEs) for modeling spacecraft dynamics. As a proof of concept, we evaluate their ability to learn accelerations in canonical astrodynamics problems, including the planar two-body and circular restricted three-body problems. We characterize the strengths, limitations, and practical caveats of the approach, assessing both accuracy and generalization across datasets of varying complexity. Results demonstrate that neural ODEs have potential as accurate, data-driven surrogates for traditional models, offering a flexible approach to modeling complex dynamics in support of improved space mission design and analysis.
Type
