Here's a schematic from the space shuttle training manuals showing how the "attitude" error signal used to drive the thrust vector controllers was determined.
"Attitude" in quote marks because it was in fact a mix of attitude error, attitude rate error, and normal acceleration error signals.
The errors would be separately gained, then filtered, then summed, then filtered again then be gained again if the crew flipped a manual switch, then be filtered again---before finally at last being sent to the thrust vector controllers.
That's a lot of filtering! And now I'm wondering: what the purpose? To remove high-frequency noise? Or to smooth the error signal again and again and again in case the preceding operation (gaining through nonlinear/discontinuous lookup tables, summing) had made the signal... insufficiently smooth?
And more importantly! What would the filters look like? Would they be low-pass filters (I want to say yes)? Would they have different cutoff frequencies? Is there an algorithm for them that I could look at (since so many space shuttle algorithms are public on the web)?