Accelerometer bias and Gyroscope bias convergence
조회 수: 40(최근 30일)
The EKF implementation sythetic Data example for Visual-Inertial Odometry https://de.mathworks.com/help/fusion/ug/visual-inertial-odometry-using-synthetic-data.html
In matlab 2018 and matlab 2020 the F matrix and G matrix produces the same results.
In model we consider and for accelerometer bias and gyroscope bias
The Kalman filter should converge to a constant bias after some time and stay constant for the rest of the trajectory. The estimated position, velocity and attitude are very good.
I couldnot explain why the gyroscope bias converges always and accelerometer bias doesnot achieve a constant value? can someone share views for a possible explaination?
Ryan Salvo 2020년 11월 17일
I'm not sure if you're looking at the R2020a or R2020b example, but for R2020b, you are seeing the accelerometer bias vary with each update since the initial state covariance is set to a low value. On the other hand, the gyroscope bias varies widely at the beginning of the trajectory since the state covariance is set to a large value.
Look at the helperInitialize function in the R2020b version of the example and you should see the following code:
% Set the gyroscope bias and visual odometry scale factor covariance to
% large values corresponding to low confidence.
filt.StateCovariance(10:12,10:12) = 1e6;
filt.StateCovariance(end) = 2e2;
You can add the following line to set the accelerometer bias covariance to obtain a plot similar to the one you have for the gyroscope bias:
% Set the accelerometer bias covariance to a large value corresponding to low confidence.
filt.StateCovariance(13:15,13:15) = 1e6;