0

I recently design my first EKF Algorithm for my drone project for attitude estimation, I am too much time confusing, about which value I put in Q and R Matrix in the real world. I use bmi088 IMu, In the IMU datasheet I see noise density, I know Q and R are covariance matrices, so I need to put their gyro and acceleration sensor's standard deviation along each axis right? how how to compute standard deviation for sensors? I see some where σ= ND * sqrt(SR )

if for example, consider a noise density of 0.01°/s/sqrt(HZ), which needs to be converted into a standard deviation at a sampling rate of 100 Hz:

σ=0.01 * sqrt(100)

so can I put these values in my process noise covariance matrix? Is this way to calculate the standard deviation for sensors like acceleration and gyro? please help me guys!

I am recenty designing EKF for attitude estimation, and I want to know which value should I put in Q and R matrices.

user3106378
  • 3
  • 1

1 Answers1

1

For measurement covariance, you would have to put σ^2 in the R matrix. For process noise covariance matrix Q, you have to make a diagonal matrix with covariances for each state. These can be tuned to achieve the best estimation.

For example, if you increase the covariances of Q, the estimation will rely more on measurements and vice versa.

kartiks77
  • 48
  • 4
  • So that question is that how to find standard deviations for IMU and GPS sensor? – user3106378 Apr 27 '23 at 09:07
  • I am not aware of the behavior of the exact sensor you are using. But a basic method is to compare the measurements against known values of the forces being applied to the system. Or your formula for extracting SD from the noise density should also work. You can also used the R matrix as a tunable parameter with your own judgement if that helps getting a better estimation. – kartiks77 Apr 27 '23 at 10:37