Robust Orbit Determination and Classification: A Learning Theoretic Approach

Srinagesh Sharma, James Cutler

Abstract:

Orbit determination involves estimation of a non-linear mapping from feature vectors associ-
ated with the position of the spacecraft to its orbital parameters. The de facto standard in
orbit determination in real-world scenarios for spacecraft has been linearized estimators such
as the extended Kalman filter. Such an estimator, while very accurate and convergent over
its linear region, is hard to generalize over arbitrary gravitational potentials and diverse sets
of measurements. It is also challenging to perform exact mathematical characterizations of
the Kalman filter performance over such general systems. Here we present a new approach
to orbit determination as a learning problem involving distribution regression and, also, for
the multiple-spacecraft scenario, a transfer learning system for classification of feature vectors
associated with spacecraft, and provide some associated analysis of such systems.