Not sure I understand the wand issue completely. That said, it is very often true that the sensor and servo functions do not need to directly address (code) the physics to achieve a reliable servo action. You can characterize the wand with some regression on the likely variables and use that to generate the input to the servo function.
Also, be vary careful screwing with the sensor data in an (often) misguided attempt to filter the data. All too often we see attempts to brutally clip (supposedly) outlier data or shortcut coded filters that generate too much noise. Do the right thing from the start with a proven effective filtering function (sinc?).