Stochastic learning and extremal-field map based autonomous guidance of low-thrust spacecraft.

A supervised stochastic learning method called the Gaussian Process Regression (GPR) is used to design an autonomous guidance law for low-thrust spacecraft. The problems considered are both of the time- and fuel-optimal regimes and a methodology based on "perturbed back-propagation" approach is presented to generate optimal control along neighboring optimal trajectories which form the extremal bundle constituting the training data-set. The use of this methodology coupled with a GPR approximation of the spacecraft control via prediction of the costate n-tuple or the primer vector respectively for time- and fuel-optimal trajectories at discrete time-steps is demonstrated to be effective in designing an autonomous guidance law using the open-loop bundle of trajectories to-go.

Direction-dependent learning approach for radial basis function networks.

Direction-dependent scaling, shaping, and rotation of Gaussian basis functions are introduced for maximal trend sensing with minimal parameter representations for input output approximation. It is shown that shaping and rotation of the radial basis functions helps in reducing the total number of function units required to approximate any given input-output data, while improving accuracy. Several alternate formulations that enforce minimal parameterization of the most general radial basis functions are presented.