Search

Several robust state estimation frameworks have been proposed over the previous decades. Underpinning all of these robust frameworks is one dubious assumption. Specifically, the assumption that an accurate a priori measurement uncertainty model can be provided. As systems become more autonomous, this assumption becomes less valid (i.e., as systems start operating in novel environments, there is no guarantee that the assumed a priori measurement uncertainty model characterizes the sensors current observation uncertainty). In an attempt to relax this assumption, a novel robust state estimation framework is proposed. The proposed framework enables robust state estimation through the iterative adaptation of the measurement uncertainty model. The adaptation of the measurement uncertainty model is granted through non-parametric clustering of the estimator' s residuals, which enables the characterization of the measurement uncertainty via a Gaussian mixture model. This Gaussian mixture model based measurement uncertainty characterization can be incorporated into any non-linear least square optimization routine. Within this dissertation, the proposed framework is instantiated into three novel robust state estimation algorithms: batch covariance estimation (BCE), batch covariance estimation over an augmented data space (BCE-AD), and incremental covariance estimation (ICE). To verify the proposed framework, three global navigation satellite system (GNSS) data sets were collected. The collected data sets provide varying levels of observation degradation to enable the characterization of the proposed algorithm on a diverse data set. Utilizing these data sets, it is shown that the proposed framework exhibits improved state estimation accuracy when compared to other robust estimation techniques when confronted with degraded data quality.