I have a Pololu 9DOF IMU that I plan to use for the Sparkfun AVC competition. One thing (of a few thousand!) that has me worried is the bias of the sensors as a function of temperature. The datasheet indicates a fairly strong effect from temperature. My first solution was to calibrate the sensors at various temperatures. But that would mean getting access to a thermal chamber. Now with the right permissions, I could have used the ones at work, but they are all big honking steel chambers. The one I am currently using is a 6 ton steel box. Not a great environment for the magnetic compasses.
So plan-B is to temperature stabilize the sensor. At first I planed to use a Thermo Electric Cooler/Heater (TEC) to do this. But talking to the vendors, I found out that they tend to have ferrous metals included. In close proximity to the magnetometers, this might not be great. So I bolted a 2 ohm 25W power resistor to a small chunk of aluminum and on the other side placed the IMU. With a PID loop I can stabilize the temperature to 110F in just a few minutes. When power is removed the temperature does not change significantly in the ~3 minutes that the competition takes. To be clear, the power must be removed from the heater to minimize the magnetic field from an active heater circuit as well as not having a 5V 3 amp power source on board. I plan on using an external power source to run the heater.
Now I need to get back to the calibration phase.
Neat ideas. My gyro appeared to suffer from temp sensitivity. Never proved it to myself but still. You are right about the many many things to think about. I will be diving in on mine full force soon.
One thing that we used successfully in our robomagellan competition with the Pololu 9DOF was to use an online estimator for the gyro bias and scale implemented with a Kalman Filter. Since you will likely already have a GPS onboard, that can give you enough observability into the system. We persisted those observations for an hour or so, such that all you needed to do was drive the car around a bit beforehand and no other calibration was necessary.
I should write up something more about that particular topic, as it seems to be one a lot of people get stuck up on.