> The systems are static ... We only see frequent software updates today because the systems are barely more then public betas.

No absolutely not. FSD models are going to need continuous updates. All real-world models exhibit "model drift". Over time most models perform less well. Real world models are monitored and retrained when needed.

Example would be when a city, state or country changes a road sign or introduces a new road sign, so a strange looking vehicle like a CyberTruck starts making an appearance on the road.

Road sign designs don't change very often. No way consumers should be forced into a perpetual monthly subscription for something that probably needs updates every 5 years or so. Not sure how it should work long-term, but a monthly subscription for a mature system sounds wrong to me.

If the FSD relies on published maps to work, it's already doomed. Road layouts change way too frequently (sometimes temporary, for construction) for that to be a requirement.

It also shouldn't really matter what kind of vehicle are out there.

If I see some weird looking car I've never seen before, I just treat it like every other car.

You might be surprised the first time, but the next time you see one you'll treat it like any other car, because you're always learning.

These systems separate the learning phase (training the model) from the acting phase, so to do the same as you they need to be able to retrain on new data occasionally as well.

Nobody's FSD can depend on map data for road signs since those can change by the hour; eg road out, construction, etc. Vision for signs is probably fine when fused with map data. But more generally, there are all kinds of road situations that need a general solution and not a map one.

Or simply the maps the car needs to rely on to find your route to your destination...