Perhaps you don't get how these things work.
No-one programs a self driving car, it learns much like we do. So if Audi/BMW drivers are a genuine threat, it will learn to recognise them as such automatically. Or more likely atm, it will learn how driving humans respond to those cars by monitoring inputs and what a reference driver does.
So what would be really interesting would be to find out whether the neural net has learnt to recognise specific cars
If the "software" is values in a neural net then the main human input is in providing new training data to improve a case where the car didn't behave well. The argument here is how big a net you need to exceed human driving ability given it is a dedicated task which makes it easier for the car, but we have magnitudes more synapses than some little Nvidia card can emulate.