[NotInControl]

Originally Posted by Jared Russell

This approach will make it appear that your robot is pitching or rolling any time you have acceleration due to any force other than gravity (such as the reactive force exerted by the carpet against the tractive force of your wheels).

That is very true... the induced acceleration due to robot motion will appear as high frequency noise on each signal. The great thing about that, is that it is easy to determine which acceleration is legitimate, and which acceleration is induced due to robot motion. Simple Low pass filters, will allow for more stable readings... but will increase delay. There is a happy medium to allow for reliable tilt measurements.

A simple Euler linear filter can be used if one desires. It is a autoregressive infinite impulse response moving average.

accelX = (accelX * alpha) + (Robot.accel.getX() * (1.0 - alpha));

Where alpha is a value from 0 to 1. as you increase alpha from 0 towards 1, the filtering gets more aggressive, and the delay increases. If alpha is 0, its easy to see that there is no filtering at all.

This is the most basic filter you can implement in my book, and it does a pretty good job. There are more robust filters if you ever wanted to experiment with them. Two popular filters to try to smooth acceleration data out would be a median filter, or a Butterworth filter. They are popular because they are relatively easy to implement in code.

If you have the means to use a Gyro to measure the axis of rotation you care about, that is the sensor I would recommend for this application... if you don't, or want to experiment otherwise, it shouldn't take too much effort to use an accelerometer to have a similar implementation.