andrewpaster
commented
7 years ago That's a good point. The video isn't clear that he's giving general equations for extended kalman filters but for the model used in this project, only radar sensors have non-linear equations.
This part of the lecture was updated with some extra info. Let me know if something is still unclear in the "EKF Algorithm Generalization" lecture.
In https://discussions.udacity.com/t/why-do-we-need-to-compute-jacobian-for-f/227492/6 we discuss lesson 5, ch 18 where Andrei says that we need to use the Jacobian for F. There is some confusion about that, because the EKF generally handles a linear motion model and where necessary, a Jacobian is used to linearise H (e.g. for handling radar measurements). The way this is worded should change to make clear that if you had a non-linear motion model, you should linearise it by using the Jacobian as you need to do for H for radar. As it stands, there is no need to linearise F for the project because the motion model is already linear.