Thanks for the update!
A further refinement would be to move the normalization to last – in other words, add the ‘fix’ logic before the normalization is done.
In case others come across this thread, you can compensate for a lot of things in software but, in general, when you have different wheels behaving differently, you should first look for other possible causes:
- check for mechanical issues, make sure they all have the same feel, etc.;
- check for electrical issues, lose connections, very uneven length wire runs with too-small gage wire, etc.;
- consider swapping various components to see if you can isolate the source of the difference as this will provide insight into what is going on;
- realize that some motors do have some bias for running faster in one direction than the other, this usually is going to be lost in the noise but it could be a factor;
- be on the lookout for non-repeatable results, such as when things shift around without explanation from test-to-test (if you find yourself chasing a moving target, take a break and then think things through again).