Derivation of Extended (Nonlinear) Kalman Filter From Scratch with Python Codes - PART I - MATH

Поделиться
HTML-код
  • Опубликовано: 23 янв 2025

Комментарии • 23

  • @aleksandarhaber
    @aleksandarhaber  Год назад +4

    It takes a significant amount of time and energy to create these free video tutorials. You can support my efforts in this way:
    - Buy me a Coffee: www.buymeacoffee.com/AleksandarHaber
    - PayPal: www.paypal.me/AleksandarHaber
    - Patreon: www.patreon.com/user?u=32080176&fan_landing=true
    - You Can also press the Thanks RUclips Dollar button

    • @RasitEvduzen
      @RasitEvduzen Год назад +1

      I finally managed to get you a coffee. 😁

    • @aleksandarhaber
      @aleksandarhaber  Год назад +2

      @@RasitEvduzen Thank you very much! It means a lot!

    • @RasitEvduzen
      @RasitEvduzen Год назад +1

      I hopes for the rapid growth of this channel, @Aleksandar Haber.

  • @micahdelaurentis6551
    @micahdelaurentis6551 3 месяца назад +1

    Thanks for this series, it's great

  • @aleksandarhaber
    @aleksandarhaber  Год назад

    Part 1 video:
    ruclips.net/video/yv3ESns38EU/видео.html
    Part 2 video:
    ruclips.net/video/qtYhKlDMww4/видео.html
    The GitHub page is given here:
    github.com/AleksandarHaber/Extended-Kalman-Implementation-in-Python
    The tutorial webpage accompanying this video lecture is given here:
    PART 1:
    aleksandarhaber.com/extended-kalman-filter-tutorial-with-disciplined-python-codes/
    PART 2:
    aleksandarhaber.com/extended-kalman-filter-tutorial-with-example-and-disciplined-python-codes-part-ii-python-codes/

  • @shaheedchikkumbi1902
    @shaheedchikkumbi1902 Год назад +1

    What method have you used for estimation of Q and R matrix

    • @aleksandarhaber
      @aleksandarhaber  Год назад +1

      We do not treat that problem in this tutorial. That is the problem for itself. We will cover this problem in our future tutorials.

    • @shaheedchikkumbi1902
      @shaheedchikkumbi1902 Год назад +1

      @@aleksandarhaber what sensor data have you used?

    • @aleksandarhaber
      @aleksandarhaber  Год назад +1

      ​@@shaheedchikkumbi1902 Watch the tutorial carefully, everything is explained well and thoroughly.

  • @robfei-u6b
    @robfei-u6b Год назад +1

    very great video, your background is control engineer?

  • @RasitEvduzen
    @RasitEvduzen Год назад +1

    Hello Aleksandar, another amazing series. in this series is it possible to explain how to tune Kalman filter. I mean how to set P,Q,R matrix. Can we tune this matris via optimization such as numerical optimization or heuristic optimization.

    • @aleksandarhaber
      @aleksandarhaber  Год назад +2

      I usually use the autocovariance approach or the auto-least squares approach. This is a data-driven approach. Here is my recent paper on this topic: opg.optica.org/oe/fulltext.cfm?uri=oe-31-11-17494&id=530519
      This paper is partly based on the work of Rawlings (references 48-49 in the paper) and other older works from 70s and 80s.

    • @RasitEvduzen
      @RasitEvduzen Год назад +1

      Sounds great, @Aleksandar Haber! I will explore your paper. Currently, I am focused on working with the Error State Kalman Filter for an INS (Inertial Navigation System) system. There are two approaches to tune the filter parameters: the traditional method and the data-driven technique. I find the filter tuning problem quite challenging because we aim to find optimal parameters for the filter within the context of stochastic processes.Finally I hope I do it.

    • @aleksandarhaber
      @aleksandarhaber  Год назад +2

      @@RasitEvduzen What do you understand under the traditional approach?

    • @RasitEvduzen
      @RasitEvduzen Год назад +1

      @@aleksandarhaber What I mean is that the first approach is based on the concept of innovation, while the second approach relies on machine learning techniques. Is it correct?

    • @aleksandarhaber
      @aleksandarhaber  Год назад +2

      @@RasitEvduzen In the paper whose link I sent you, I use a modified approach of Rawlings. That is the optimization approach that uses the innovation sequence and correlations to find the matrices Q and R. I will make a video about this approach and post my MATLAB codes on GitHub, so you can take a look. I performed significant testing. I am not very satisfied with this approach, to be honest, since the results seem to be biased. However, we can still get some initial estimates of Q and R. This is the optimization approach. In my understanding, the more traditional approaches are based on manual tuning through trial and error. It is known that you can start with a rough model and use the matrix Q to tune the model. Anyway, I will look more into this. ALSO, THANK YOU FOR THE SUPPORT AND COFFEE!