Title:Generalized Moving Horizon Estimation for Nonlinear Systems with Robustness to Measurement Outliers

Abstract:Moving horizon estimation (MHE) is an effective filtering technique for dynamic systems. While there has been noticeable progress in the stability analysis of MHE, there is lack of research on robustifying MHE against measurement outliers. To bridge this gap, we propose a generalized MHE approach by utilizing the loss-theoretic perspective of Generalized Bayesian Inference. In particular, we design a robust loss function by leveraging the \{beta}-divergence and propose the \{beta} moving horizon estimator to handle the outliers. Analytical influence functions are derived to analyze the robustness of the MHE methods. Based on this, we prove that for the case of linear Gaussian systems, the gross error sensitivity of the proposed estimator remains bounded, while for the standard MHE, it is unbounded. The effectiveness of the proposed approach is demonstrated in simulations on both linear and nonlinear systems.