8i20

One not-very-scientific way to get the encoder offset is to just guess it until the motor runs at the same speed in both directions with a low drive current.

The u+ VW- method is the best way, but it's not so easy to do that with an 8i20 as it adds in a phase angle.

If you can connect the servo to a power supply on the bench, then your idea will work. Set the "index-enable" of the encoder counter, turn it by hand through the reset, and then look at the encoder counts and take the modulus, as you suggest.

It is likely to be zero, or very close to zero.

Why are my outputs all "spikey" ?

What does your rotor position feedback look like? (probably encoder.0.position)
Also look at the encoder velocity.

So I reduced the encoder counts from 8192 (which were clean - not spikey) to 1536 ppr, because I doubt 8192 x 3000rpm on three axis is gonna be kept up with anyway...

Still getting a non-smooth motion in two dimensions - tiny reactions to encoder inputs and some kind of long sinusoidal that I guess is from the motor cogging on 80% power.... Is the pid not smoothing though ? or is my I too low?
 

High speed pass.
Is the pid overreacting for some reason ?
 
 

Hmm. Try zeroing out the D term?