1
$\begingroup$

I do have a land-based robot with an IMU and a GNSS receiver.

  • From the IMU, I get the velocity and acceleration in both $x$ and $y$ directions.

  • From the GNSS receiver, I get the latitude and longitude.

I want to fuse these data to get my exact position on the field. To get the most accurate position I want to use an Unscented Kalman Filter (UKF).

However, I don't really understand the concept of fusing these data. Most of the books I found just fused the IMU data and used it together with the GNSS data but by my understanding, I should get a more precise position when I fuse IMU and GNSS.

Can someone explain to me the concept of doing so or has a good source/tutorial?

This is for my bachelor thesis. Help would be much appreciated. Thanks!

$\endgroup$
0
$\begingroup$

How did you come to the conclusion that you will need an UKF? Make sure you understand the math behind a Kalman Filter first and understand why you would need an EKF or UKF over a normal KF. This article describes the math behind a Kalman Filter using an IMU but you can add more sensors to this setup.. https://www.bzarg.com/p/how-a-kalman-filter-works-in-pictures/

Also, what platform (hardware/software) do you plan to implement your project in?

| improve this answer | |
$\endgroup$
  • 1
    $\begingroup$ The Robot is driving of-roads on predefined lanes with different turns after every lane + he may need to stop for other robots and so on. While all of this I need cm-accuracy at all time that's why I went for an UKF as its using a better way of linearizing the non linear functions. My platform is a embedded system with linux running, programming will be in C or C++. Thanks for the link! $\endgroup$ – Strohhut Sep 6 at 7:57
0
$\begingroup$

I cannot recommend the robot_localization package in ROS enough. Please see my response to another post I made today How does sensor fusion help in robot localization. The documentation for the package is superb and I think, assuming you have ROS avaliable, you can have a EKF or UKF up and running in a week.

And as it regards the authors claim in a comment to another answer that the UKF provides "a better way of linearizing the non linear functions" I think it should be clear that is not true. A UKF does not linearize anything. It just picks some well placed samples (called sigma points) and passes them through the nonlinear functions, calculating weighted approximations to the expected values and covariances as it goes along. It is not calculating a Jacobian and linearizing about the mean like the EKF does.

| improve this answer | |
$\endgroup$

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.