i'd just call it a heuristic feedback system.

basically you're recognizing three magnitude ranges of offset and assigning one response correction to each range, looping until converges upon the minimum tracking offset.

i wouldn't call that fuzzy because there is no randomness to the feedback response. its a one to one assignment, a straight mapping.

right now your algorithm is stepwise; if you want to consider speeding it up (if the potential offsets are great and the speed to converge important) then you could track the error between subsequent feedback tests instead of just the offset:

if your range test was the max and then the max again, you could bump the feedback correction a little more. this gives you a little non-linearity which could speed things up on large offsets though you need to be mindful that it will/can overshoot and converge like a damped oscillation. effectively it would be a small step closer to PID implementation.

PID in my opinion is nothing magic or cosmic. its just a more sophisticated convergence algorithm than programmers would normally want to program.

a lot of applications gain with that extra difference. but technically even PID is a heuristic approach that is just measured by its resulting improvement.

duh