ContextQMD
Back to Openmower

Magnetometer Calibration

utils/mag_calibration/README.md

0.13.22.3 KB
Original Source

Magnetometer Calibration

In an ideal word, when we rotate our robot around the Z axis, our magnetometer values would create a circle around the origin. Unfortunately, in the real world, due to magnetic fields in the robot and sensor bias, this is not the case. The magnetometer will most likely be an ellipse or a circle with some offset from the origin.

This calibration will remove this offset, but won't remove any elliptical form!

How I'm currently doing this

Currently, I am doing it in the following way:

  1. Put the robot in your lawn (turned on, fully assembled)
  2. Start a teleop + recording session. Rotate the robot slowly in one spot for several rotations.
  3. Run the plot_mag.sh script: ./plot_mag.sh /path/to/your.bag
  4. Check in the interactive plot, if your values are circular and if the fitted circle (red circle) is a good fit.
  5. If everything seems fine, your bias values will be printed in the console (see example below).

Example

This is the output of my calibration run today.

Note that:

  1. The outliers have been rejected
  2. The plot is pretty much a circle, moved to the upper right hand side

The console output was the following:

bash
[...]

After 3 iterations the fit converged.
final sum of squares of residuals : 8.48886e+07
rel. change during last iteration : -4.36383e-06

degrees of freedom    (FIT_NDF)                        : 16540
rms of residuals      (FIT_STDFIT) = sqrt(WSSR/ndf)    : 71.6402
variance of residuals (reduced chisquare) = WSSR/ndf   : 5132.32

Final set of parameters            Asymptotic Standard Error
=======================            ==========================
bx              = 11.4452          +/- 0.0209       (0.1826%)
by              = 12.2336          +/- 0.02425      (0.1982%)
r               = 370.698          +/- 0.5595       (0.1509%)

correlation matrix of the fit parameters:
                bx     by     r      
bx              1.000 
by             -0.402  1.000 
r               0.092 -0.020  1.000

--> Our bias parameters (bx, by) are 11.4452 and 12.2336

Better Calibration

This assumes that the result will be a circle with an offset. In reality, the result could be an ellipse. Also it would be nice to automate this process.