Near-linear uncertainty quantification and tracking in the cislunar regime
Abstract
Cislunar Space Domain Awareness (SDA) is extremely challenging due to uncertainty distributions that rapidly become non-Gaussian and extend across large regions of the state space. State-of-the-art UQ methods rely on the decomposition of uncertainty distributions on Gaussian or polynomial bases, which results in increases in realism at the cost of significantly worsened computational efficiency. We propose a paradigm shift in UQ by developing mathematical representations of the three-body problem in which the uncertainty propagation is well-behaved. Regularizations of the three-body problem are employed to globally rectify the dynamics and thus obtain a well-behaved representation of the system. This is applied for the first time to enhance non-intrusive UQ methods and develop a new regularized filtering scheme in which uncertainties grow quasilinearly. Building on the gained insight, we develop a new scientific machine learning method to automatically discover coordinate transformations in which UQ is efficient. Regularizations and automatically discovered coordinates have enjoyed considerable success in rectifying dynamical systems across several engineering domains. The complexity of cislunar UQ under severe non-linearity is tackled through an incremental approach which allows us to gradually build insight into how uncertainty diffuses through the 3-body problem. The goal of the proposed research is to build a UQ method that will form the basis for future cislunar catalogue building and maintenance, ensuring the stability and security of the cislunar regime for future generations.
Document Details
- Document Type
- DoD Grant Award
- Publication Date
- Feb 05, 2025
- Source ID
- FA86552417044
Entities
People
- Davide Amato
Organizations
- Air Force Office of Scientific Research
- Imperial College London
- United States Air Force