2024-09-16, 03:50 PM
In my experience with Kalman filters in other projects
without biases as states variables in the model the Kalman filter simply converges to
values with error (basically wrong , skewed values).
So having biases and estimating them in Kalman process is essential.
Using Kalman model without estimating biases and relying on sensor self correction leads
to too big systemic errors.
There papers which often publish some Kalman model which assume no bias (zero mean noise)
People code them as is, without extending matrices for biases. These filters do not work with real sensors.
I could never make compass in pypilot work without unreasonable drift.
I’ve done all calibrations. There were fine. I do cicrcles on a boat after leaving marina to give it
chances to auto calibrate. But the drift is too high. And I think it’s due to RTImuLib2 Kalman
not having biases in the dynamic model.
Thanks
without biases as states variables in the model the Kalman filter simply converges to
values with error (basically wrong , skewed values).
So having biases and estimating them in Kalman process is essential.
Using Kalman model without estimating biases and relying on sensor self correction leads
to too big systemic errors.
There papers which often publish some Kalman model which assume no bias (zero mean noise)
People code them as is, without extending matrices for biases. These filters do not work with real sensors.
I could never make compass in pypilot work without unreasonable drift.
I’ve done all calibrations. There were fine. I do cicrcles on a boat after leaving marina to give it
chances to auto calibrate. But the drift is too high. And I think it’s due to RTImuLib2 Kalman
not having biases in the dynamic model.
Thanks
Download BBN Marine OS for raspberry pi
https://bareboat-necessities.github.io/m...at-os.html
Video of actual installation:
https://www.youtube.com/watch?v=3zMjUs2X3qU
https://bareboat-necessities.github.io/m...at-os.html
Video of actual installation:
https://www.youtube.com/watch?v=3zMjUs2X3qU