Skip to main content

ORIGINAL RESEARCH article

Front. Neurorobot.
Volume 18 - 2024 | doi: 10.3389/fnbot.2024.1460255
This article is part of the Research Topic Advancing Neural Network-Based Intelligent Algorithms in Robotics: Challenges, Solutions, and Future Perspectives View all 14 articles

KalmanFormer: Using Transformer to Model the Kalman Gain in Kalman Filters

Provisionally accepted
Siyuan Shen Siyuan Shen 1Jichen Chen Jichen Chen 2*Guanfeng Yu Guanfeng Yu 3*Zhengjun Zhai Zhengjun Zhai 1*Pujie Han Pujie Han 4*
  • 1 Northwestern Polytechnical University, Xi'an, China
  • 2 Xi'an Microelectronics Technology Institute, Xi'an, China
  • 3 AVIC Xi'an Aeronautics Computing Technique Research Institute, Xi'an, China
  • 4 Zhengzhou University of Light Industry, Zhengzhou, Henan Province, China

The final, formatted version of the article will be published soon.

    It is a fundamental task for tracking the hidden state of the dynamic systems in signal processing.Recursive Kalman filters (KF) are an excellent option with low complexity when it comes to fully known linear and Gaussian systems. Nevertheless, it can be challenging to meet the requirements of non-linearity when applying it to real-world applications. Furthermore, it is difficult to accurately describe the model and noise of practical dynamic systems. This paper presents the KalmanFormer, a hybrid model-driven and data-driven state estimator that learns from data to improve the performance of Kalman Filtering under nonlinear conditions with partial known information. We employ a Transformer framework to track the Kalman Gain from the data without prior noise parameters and we combine the learned Kalman Gain into the classical data flow of the Kalman Filter. Numerical experiments demonstrate that the KalmanFormer overcomes non-linearities and model mismatch, outperforming the classic Extended Kalman Filter operating in the same condition.

    Keywords: Kalman filter, deep learning, transformer, Kalman gain, supervised paradigm

    Received: 05 Jul 2024; Accepted: 10 Dec 2024.

    Copyright: © 2024 Shen, Chen, Yu, Zhai and Han. This is an open-access article distributed under the terms of the Creative Commons Attribution License (CC BY). The use, distribution or reproduction in other forums is permitted, provided the original author(s) or licensor are credited and that the original publication in this journal is cited, in accordance with accepted academic practice. No use, distribution or reproduction is permitted which does not comply with these terms.

    * Correspondence:
    Jichen Chen, Xi'an Microelectronics Technology Institute, Xi'an, China
    Guanfeng Yu, AVIC Xi'an Aeronautics Computing Technique Research Institute, Xi'an, China
    Zhengjun Zhai, Northwestern Polytechnical University, Xi'an, China
    Pujie Han, Zhengzhou University of Light Industry, Zhengzhou, 450002, Henan Province, China

    Disclaimer: All claims expressed in this article are solely those of the authors and do not necessarily represent those of their affiliated organizations, or those of the publisher, the editors and the reviewers. Any product that may be evaluated in this article or claim that may be made by its manufacturer is not guaranteed or endorsed by the publisher.