Quote:
Originally Posted by ekapalka
That's pretty much the conclusion I arrived at. There must be some correlation between rotating and accelerometer output, though, like centripetal acceleration?
|
An interesting experiment is to put a smartphone on a flat surface and rotate it. Usually the screen display doesn't change between portrait/landscape unlike typical usage when the phone is tilted. This is also because the accelerometer can't distinguish orientation in the X-Y plane (yaw).
The key advantage of the complementary filter is that it's far simpler than approaches like the Kalman filter if you make certain assumptions about the signals. So in the case of an aerial vehicle we can assume that total accelerometer values near 1g represent a stable orientation.
If you have wheel encoders on a wheeled robot you can make similar assumptions about rotation based on the left-right count difference.