[eugenebrooks]

Kevin's code implements a deadband. Although this is just fine
for using it to measure a turn, it is not so good for using the gyro
as a compass.

Using the gyro as a compass requires you to be entirely linear
with no deadband, and to carefully average the "no turn" output
(noise helps you here) and subtract the value using fixed binary
point arithmetic. Second order accurate time integration is also
a good idea in this case.

A drift of a couple of tenths of a degree during the 15 second
autonomous mode is achievable, but the drift is sensitive to
temperature and you either need to recalibrate, or compensate,
for that. What we do is check the drift now and then, and then
recompute the required average when the drift has picked up
due to temperature change.


Eugene