Multi-Fidelity Orbit Determination with Systematic Errors

Multi-fidelity approaches to orbit-state probability density prediction reduce computation time, but introduce a systematic error in the single-point prediction of a spacecraft state. An estimate of the systematic error may be quantified using cross-validation. Credibilistic filters based on Outer Probability Measures (OPMs) enable a principled and unified representation of random and systematic errors in object tracking. The quantified error of the multi-fidelity approach defines an OPM-based transition kernel, which is used in a credibilistic filter to account for the systematic error in the orbit determination process. An approach based on automatic domain splitting is proposed to reduce the error beyond what is normally achievable with multi-fidelity methods. A proof-of-concept for the approach is demonstrated for a simulated scenario tracking a newly-detected space object in low-Earth orbit via two ground stations generating radar measurements. An OPM-based definition of the admissible region combined with the multi-fidelity credibilistic filter establishes custody of the object.