In a state estimation scheme with a simple IMU/GNSS setup using EKF, I have always thought the prediction step would be done using a motion model and all sensor measurements would be incorporated via observation updates.

However i have noticed IMU measurements being used for prediction in several works using something called a strapdown inertial navigation (Example) which is used as the prediction model with no mention to a kinematic motion model. Also in this paper i see that the prediction step is done by integrating the IMU measurements with the motion model.

This is slightly confusing with my concept of temporal and measurement update duality. Is strapdown inertial navigation and using a kinematic model are substitutes for each other or are they complementary? Do you know of any works that compare them in case they are analogous? Any good reads to understand these concepts would be appreciated.

  • $\begingroup$ Can you give an example using kinematic model? I've only used (with good success) the method that uses the IMU outputs as inputs to an unconstrained motion model with position outputs, then used those estimated outputs with GPS to estimate states. $\endgroup$
    – TimWescott
    Commented Jun 28, 2019 at 1:02
  • $\begingroup$ @TimWescott Something like a constant acceleration model where you know hot to transform the state for a given time difference. As in when a GPS measurement is received, a prediction would be made using the previous state and the time delta since the last update. $\endgroup$
    – vasf
    Commented Jun 28, 2019 at 14:24
  • $\begingroup$ Eh, I was hoping for a link or links to papers. The only way I could see to make it work was the strapdown method -- which works astonishingly well, by the way. $\endgroup$
    – TimWescott
    Commented Jun 28, 2019 at 17:51


Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.