Seminars

In state estimation, quantities of interest, referred to as states, are determined using available information. This information usually comes in the form of measurements, models, and noise statistics. State estimation can be approached in different ways. This dissertation focuses on filtering and smoothing. In filtering, all past and present information is used to estimate the current state. In smoothing, all past and present information is used to estimate one or more past states. Within this context, three main contributions are presented in this work. The first contribution addresses smoothing in a factorized framework. Two fixed interval smoothers are developed in a UDU framework by applying low rank updates and introducing the weighted hyperbolic Householder reflector. The second contribution presents an adaptive framework for atmospheric entry. In this framework, atmospheric density is estimated with an adaptive neural network, which improves navigation and guidance performance. The third and final contribution focuses on improving the nonlinear filter known as the point mass filter (PMF). In this contribution, the design of the PMF is improved by centering the grid of point particles at the approximated posterior distribution rather than at the prior. In addition, a new technique for optimally sampling grid points is introduced, resulting in more robust estimation performance.