In general descriptions of the Kalman Filter it is described as "fusing" of sensor data.

Some sensors (e.g. wheel rotation counters, gyros, accelerometers) are considered as input to the Prediction step (odometry) whereas others (rangefinders, GPS) are input to the Measurement step.

My question is does the Kalman Filter treat the sensors in these two categories any differently?

If one (of three) gyros gives bad data will the Kalman Filter start to ignore the readings from that gyro?

How is that achieved?

For the Measurement Step clearly the Kalman Gain can start to reduce the column values corresponding to a particular sensor but note that those are the sensors that provide data for the Measurement Step.

So how does Kalman reduce the impact of a bad odometry sensor? Will the Kalman-gain weight the measurement data higher in that case?

Clearly the sensors in these two groups (Prediction, Measurement) cross-check each other but how about sensors from within a group? If we have redundant information within the Odometry sensors (e.g. speedometer and accelerometer) will the Kalman Filter arbitrate appropriately between the two?

What are the design considerations for assigning a sensor to the Prediction step or Measurement step or is it obvious which goes where?