Perhaps this post will help:
http://www.chiefdelphi.com/forums/sh...65&postcount=6
It might also help to figure out if your code potentially sends a 0 instead of 255 based upon joystick input to motor control algorithms. Excel or Matlab would be great to detect this since you could map out expected motor outputs based upon the joystick input map.
The 'filtering' Russ refers to is as simple as 'smoothing' the outputs to the motors by not allowing two chronological outputs to have a difference greater than N, where N is determined experimentally. Typically we found that if we gave it 300-500ms in between full forward/reverse this problem never came up, yet that was also on a slick field.