[uutzinger]

I would be interested in the performance of these self made units.
In particular I would like to know what the magnitude is of the following measurements when the sensor is not moving.

Gyration in radians/sec
Acceleration repeatability in g (turn off / turn on again after a few days)
Magnetometer variation in % at run time

I am working with RTIMU2-teensy and have ported code on my github account (uutzinger). I am using the MPU 9250 which is the same as the navx mxp unit. The RTIMU library supports many other IMUs including the Bosh unit. RTIMU worked for me because I could expand the code to make an underwater pressure sensor work with it.

With the MPU 9250 i get
Gyration 0.002 rad/sec when sensor does not move.
Acceleration is within 10mg after a few turn on/off cycles.
Magnetometer varies usually about 10-20% depending on orientation even after calibration. The magnetomter is the main source of fluctuations in the pose.

I believe the key ingredients for IMU to work well on an FRC robot is that you can turn off the magnetometer when motors are engaged and that you can accurately measure the gyroscope bias when you know the robot is not moving.

The navx unit has an magnetic field anomaly detection but I have not yet seen that working correctly and I think its better for teams to turn off the compass when their autonomous period starts (and navigate with gyroscope and accelerometer).

RTIMU computes the gyro bias during runtime which has its own issues because detecting that the IMU is not moving is tricky.

Even with best calibration the gravity removed acceleration I get is about 0.02 m/s2 when the robot does not move.