Utumno
2 x Model Y LR (2022 Q2 & 2023 Q3)
The wiper response model is trained en-masse using data from the fleet, as you said. The inference side of it is, of course, vehicle specific and runs on the vehicle's computational power. That inference result decides the wiper response to real-time vehicle data inputs.Nothing is learned from individual cars. Its all fleet aka crowd sourced. When it first came out, called deep rain, they did say that they took inputs from the cars if you hit stalk to override but it’s been a while since I have seen any evidence of data from our car being download for any aspect. In early days it was a common occurrence, often after most drives. I guess Tesla have more data sources than they need now.
It's therefore perfectly possible for two vehicles to have sufficient disparity between those inputs to infer two completely different results from the same wiper response model.