Sorry, no Kalman filter script that I know of. Accelerometer data (the tilt values) is usually pretty solid, unless you are rolling over a really rough surface. The gyro data (the angular velocity values) is very spiky by nature, unless your robot is in the air gyro data isn't all that beneficial. You could measure impact or freefall with the gyro data I guess. The compass data (direction) is a bit jumpy at times, but compass data isn't something that you poll for continuously.

What's your application? Aerial or ground?

Thanks Jeremie. I was attempting to use it on Roli and continuously poll it for compass heading. Sitting still the signal is reasonably stable but rolling across the floor causes huge data variances and drift in signal. I thought initially that I could find a correlation between Roli rolling along and Acceleration. I was hoping to apply a correction to Compass and correct for drift. But alas it was not to be. Yet. I suppose I need a different strategy.
Jack

Kalman or complimentary filters are math. Which means you can write your own in ezscript. Or, if you find one filter based on c#, you can create a simple plugin. Have fun!