Abstract:Practical Bayes filters often assume the state distribution of each time step to be Gaussian for computational tractability, resulting in the so-called Gaussian filters. When facing nonlinear systems, Gaussian filters such as extended Kalman filter (EKF) or unscented Kalman filter (UKF) typically rely on certain linearization techniques, which can introduce large estimation errors. To address this issue, this paper reconstructs the prediction and update steps of Gaussian filtering as solutions to two distinct optimization problems, whose optimal conditions are found to have analytical forms from Stein's lemma. It is observed that the stationary point for the prediction step requires calculating the first two moments of the prior distribution, which is equivalent to that step in existing moment-matching filters. In the update step, instead of linearizing the model to approximate the stationary points, we propose an iterative approach to directly minimize the update step's objective to avoid linearization errors. For the purpose of performing the steepest descent on the Gaussian manifold, we derive its natural gradient that leverages Fisher information matrix to adjust the gradient direction, accounting for the curvature of the parameter space. Combining this update step with moment matching in the prediction step, we introduce a new iterative filter for nonlinear systems called \textit{N}atural Gr\textit{a}dient Gaussia\textit{n} Appr\textit{o}ximation filter, or NANO filter for short. We prove that NANO filter locally converges to the optimal Gaussian approximation at each time step. Furthermore, the estimation error is proven exponentially bounded for nearly linear measurement equation and low noise levels through constructing a supermartingale-like property across consecutive time steps.

Nonlinear Estimation by Linear Estimation with Augmentation of Uncorrelated Conversion.

Parameter Estimation in Exponential Models by Linear and Nonlinear Fitting Methods.

A Unitary Transform Based Generalized Approximate Message Passing

Augmented Nonlinear Least Squares Estimation With Applications to Localization

Multi-sensor Optimal Linear Estimation with Unobservable Measurement Losses

Maximum correntropy unscented filter based on unbiased minimum-variance estimation for a class of nonlinear systems

Robust estimation in a nonlinear cointegration model

A Novel Fractional Nonlinear State Estimation Algorithm in Non-Gaussian Noise Environment

A Nonlinear Stochastic Filter for Continuous-Time State Estimation

Measurement-driven Quality Assessment of Nonlinear Systems by Exponential Replacement

Nonlinear Bayesian Filtering with Natural Gradient Gaussian Approximation

A new method for the nonlinear transformation of means and covariances in filters and estimators

Distributed Nonlinear Fusion Estimation for Networked Multi Sensor System with Noisy Control Input and Packet Dropout

Discrete-Valued Signal Estimation via Low-Complexity Message Passing Algorithm for Highly Correlated Measurements

Robust Cubature Kalman Filter With Gaussian-Multivariate Laplacian Mixture Distribution and Partial Variational Bayesian Method

Robust Fractional Nonlinear State Estimation Against Random Incomplete Measurements and Unknown Noise Statistics

Unlabeled Compressed Sensing from Multiple Measurement Vectors

An improved Bayesian filter for nonlinear systems under multistep randomly delayed and lost measurements

Distributed Nonlinear Fusion Estimation Without Knowledge of Noise Statistical Information: A Robust Design Approach

Robust sequential fusion Kalman estimators with asymptotic equivalence and stability for networked uncertain sensor systems

Robust Power System Dynamic State Estimator with Non-Gaussian Measurement Noise: Part II--Implementation and Results