Making regular GPS ultra-precise

The social norm that we seem to be slowly stumbling into assumes a different state of affairs:

That the roadways ought to always be clear of anything but cars which are behaving normally, and that if you can keep a car in the lane then anything unusual which happens isn't the car driver's fault.

Debris, animals, children, pedestrians, cyclists, motorcycles, stopped cars, construction workers and their vehicles, and anything else in our complex world that may find its way into the road - these aren't visible to someone blinded by oncoming headlights at night, they don't register in the same way as a car in the peripheral vision of someone looking at their cell phone or touchscreen controls, they may not be detected by radar cruise sensors or lane centering cameras, they certainly won't register on a GPS navigation track...and in a collision, society increasingly blames the thing that wasn't ~~supposed to be there~~ *anticipated* to be there rather than the driver which crashed into the thing.

I recognize that the likely cause of this is simply the infrequency of those events. Spend a few thousand miles seeing little other than cars on the road, and it's easier for your brain to assume that cars are the only thing that can be on the road. But my skeptical, cynical, conspiracy-minded side wonders if some of this trend is encouraged by submarine marketing efforts from self-driving vendors - the problem gets a lot easier when your "autonomous" vehicle isn't at fault for hitting a pedestrian in the road and you can just follow a GPS track while sensing for 5000 lbs steel boxes following the same GPS track.

There's no need with hybrid positioning in cellular location chipsets that use cell towers, one or more GNSS constellations, and/or a WiFi SSID GIS database (like Skyhook).

Why over a special radio system? Why not over the internet?

That's not so good. 24-channel GPS can do while moving (that's RTK) using (free in most areas) WAAS. Fusion constellation many channel GNSS receivers with clear terrain can do about as good without DGPS (of which RTK is a subset). There's really no reason not to use as many constellations and as many satellites as possible to reduce error more quickly.

I really hope QPS works, becomes available to the public, and decreases in cost quickly because it would make GNSSes more of a backup and calibration system than a primary system. Hybrid location chipsets in mobile handsets use a fusion of many technologies and techniques which would be really cool to have QPS especially for underground metro systems and inside buildings with lots of signal-attenuating materials.

So you can deploy a bazillion devices without each one needing internet access, which typically has a cost because there's administrative overhead to handling abuse complaints, etc, that come along with full internet access.

If all you got was a one-way stream of data with no internet access (assume your CORS virtual station location is the AP's location, so no need to even pick your own mountpoint) then the abuse potential is basically nil. Just like with FM RDS and stuff, it's just a broadcast that you receive, you can't do anything bad with it. It's not internet, it's just one-way data. Difference being that a lot of microcontrollers now have wifi MAC/PHY built in, whereas FM would need more silicon.