Honestly, right now it doesn't matter. Not even at AP1. Some people think vehicle taxonomy translates to more "confidence" in what the car is perceiving, but IMO that's just human bias. I don't care if it sees what kind of vehicle it is or what color it is or how many passengers it has. I care that it's detecting distance, speed, and the bounding box for the object, and that has nothing to do with what graphical asset is chosen on the IC.
Now, if AP would keep away from a truck, or lane-split for a motorcycle, or otherwise know to avoid nearby vehicles based off their behavior or appearance, then I'd start caring. But so far, I've not even seen AP1 do that (it's more liberal at lane-offset based off ultrasound detection, but not based off vision)