[RyanW]

What Kaszeta said is correct; however, our team decided we also wanted to scale the values. What we came up with was to take the initial inputs, change the range to (-127)-(127), add them together, then scale the sum by a factor:

The factor = Max ( |speed|, |turn|, |strafe|) / (|speed| + |turn| + |strafe|)

We used this code last year, and it seemed to work pretty well.