Sensors!

[Jared Russell]

I have done quite a bit of work with gyros and drift is one of the more difficult things to characterize. What we call "drift" is the effect of integrating the output of a noisy signal. In the case of a MEMS gyroscope, this noise is caused by small thermal variations and vibrations that are impossible to completely correct for. Imagine a variable with noise of +/- 1 unit at any given time. What happens if you integrate this variable? You get what is called a "random walk", or drift.

Because electromechanical noise is usually modeled as a random variable (frequently as white noise with a mean and a variance, but no single exact value), you can never say for certain what your worst case drift will be. But, you can calculate things like your "average" drift rate over time, or your drift over various confidence intervals.

Drift rate is affected by several factors, including:
* The gyroscope's noise properties
* Sampling rate
* Temperature
* Filtering