Loading [MathJax]/extensions/MathMenu.js
EKFNet: Learning System Noise Statistics from Measurement Data | IEEE Conference Publication | IEEE Xplore

EKFNet: Learning System Noise Statistics from Measurement Data


Abstract:

In this paper, to reduce the time and manpower spent on fine-tuning an extended Kalman filter (EKF), we propose a new learning framework, EKFNet, for automatically estima...Show More

Abstract:

In this paper, to reduce the time and manpower spent on fine-tuning an extended Kalman filter (EKF), we propose a new learning framework, EKFNet, for automatically estimating the best process and measurement noise covariance pair from the real measurement data. The EKFNet is trained by using backpropagation through time (BPTT). The proposed method can choose among several optimization criteria, such as maximizing the likelihood, minimizing the measurement residual error, or minimizing the posterior state estimation error. We illustrate the proposed method's performance using real GPS data, which outperforms existing methods and a manually tuned EKF.
Date of Conference: 06-11 June 2021
Date Added to IEEE Xplore: 13 May 2021
ISBN Information:

ISSN Information:

Conference Location: Toronto, ON, Canada

1. INTRODUCTION

Popular nonlinear filters for state estimation include the extended Kalman filter (EKF) [1], [2], the unscented Kalman filter (UKF) [3], and the particle filter (PF) [4]. In practice, the EKF is widely used due to its much lower computational load than the PF and UKF. In the motion and measurement models of a non-linear filter, noise terms are used for compensate the errors made by: (1) model simplification, (2) additional states not modeled, (3) discretization error, (4) model linearization, and (5) Gaussian noise assumptions. However, it takes significant amount of time and man power to manually fine-tune the EFK and acquire proper system noise statistics used by the filter. It is thus desirable to develop an approach that automatically finds best noise statistics for an EKF.

Contact IEEE to Subscribe

References

References is not available for this document.