Wheel odometry : hundreds of tests

I started this robot session by "re-centering" the servos : no rotation for pulses duration = 750. A very basic and mandatory tuning. I ran the calibration routine "calibrate_all.bs2", and this time, I could take advantage of the sun light that is supposed to give better results because of the photoresistor. Let's assume that the term "sun light" is appropriate for this cloudy sunday afternoon in Paris...
Today's value of SDIRINC = $719C
This is similar to the previous calibrations : $70F9, $7211, $71AC...

Okay... Then I spent a good amount of time running test programs, slightly adjusting the calibration values and tuning the "wheel_odometry.bs2" program.
Each run is recorded using the wheel encoders, and these coordinates are imported into Excel to compare the record to the real path :

I managed to find some acceptable SDIRINC values, but they seem to be "dedicated" to a type of motion :
- $6A40 for continuous motion (rotation using the 2 servos at a different speed)
- $7000 for "discrete" motion (rotation using one servo stopped and one servo at full speed)...

Obviously, the wheel encoders require some extra work before this endless project can move to the next step !