MEMS IMU Calibration

[reph]

What is the state-of-the-art is for calibrating cheap MEMS IMUs (compensating for bias, non-linearity, temperature drift, variations and crosstalk between axes, etc)?

I'm thinking about a stepper motor based sled and a rotating plate. Both with optical encoders to track exact position vs time. SW would run through different motions, compare the IMU data vs the encoders, and generate correction factors to minimize error.

Surely there's something better than just tuning constants until the plane stops crashing?

[drex]

kalman filtering

:D :D

[rbeall]

kalman filtering with gps and temperature compensation

[reph]

Then, how do you generate the constants for the kalman filter? :p
I don't see how it can work without bench tuning/calibration..

[rbeall]

I just tuned the kalman gains on the bench and then readjusted in flight until everything looked about right.

[KeiserSose]

From my own experience:

Accelerometers calibration is the simplest. Can be done initially by fixing IMU in different positions to collect average data (at least 9 poitions , because you want to find 3x3 triangular conversion matrix and 3 offsets). Next, solve system of non-linear equations to find offsets and conversion matrix. This is based on the fact that you should see a sphere with 1g radius. It will also correct non-ortogonality. If non-linearity observed, I would guess, you could adjust it later, when initial non-ortogonality and bias corrected.

Magnetometers. Similar approach due to permament and constant magnetic field. Correction on geographical location, if important, can be done trivially knowing actual coordinates.

Gyros. We used the same apporach, as you proposed. "Ideal" encoders-based IMU coupled with unit being calibrated. In theory, same approach as above, but using encoders reference data. There are also commercial services that do gyros calibration precisely. But we never got that far.