search

KumarRobotics / msckf_vio

Robust Stereo Visual Inertial Odometry for Fast Autonomous Flight
Other
1.71k stars 599 forks source link

EKF update confusion #74

Closed fushi219 closed 5 years ago

fushi219 commented 5 years ago

Hello, thanks for your great job. I got a question about the MsckfVio::measurementUpdate function. When it comes to update the error state,which has been done in the line-966:

// Compute the error of the state. VectorXd delta_x = K * r_thin;

I'm confused that the calculated delta_x here seems to be just the correction part of the error state X ,which means that the error state should be "X+delta_x" after EKF update. Then I could use the new error state(X+delta_x) to correct my nominal state. If so ,why the code here only use the correction part of error state (delta_x) to do correction of nominal state ?Where is my original error state X ?

ke-sun commented 5 years ago

The mean of the error state is always 0 in the prior distribution. Therefore, whatever delta_x is, it is the updated mean of the error state, which is then directly used to update the nominal state.

fushi219 commented 5 years ago

I got it，thank you so much！ :-)