Anyone successfully using the accelerometer to derive(integrate) the robot speed (and direction given it has multi-axis support) ? Any hints about filtering ? How far off is its speed estimates ?

Any code around anywhere for it (other than the basic class in WPILib) ?

Any idea how many "G"s we can expect in robot bumper based collisions ? Could the max “G” be greater than the 3 Gs the sensor supports and hence the resultant speed would be essentially corrupted after a bigt collision ?

We did something like this experimentally. We ran a Riemann sum by using the numerical data from the accelerometer and the delay from our PID loop. It’s defiantly not exact, and becomes worse if your trying to get position as well, but how much depends on the acceleration of the robot(derivative of the jerk function, concavity of the acceleration). We simply did it to see if it can be done. We never tested it to see how far off it would be. It would be interesting to see though.

I can’t say how many G’s will be on the robot in bumper-bumper collisions, mainly because that depends on speed(Momentum), and the length of impact(Force=integral of momentum with respect to time).