The mobile phone industry has been trying to sell this for years. So far they have not had much success with it. The problem is that the existing infrastructure is not really sufficient. The first problem is coverage. There are plenty of roads and intersections without adequate cell phone coverage and what would the cars do when they reach those? Sure, one could put a picocell at every intersection and the mobile phone industry would love to do that - if someone paid for it. Second problem is reliability. A typical cell is available like 99% of the time, which may sound high but means that it is down 3 days in a year. Third problem is latency which I'll address below.
Having dedicated vehicle-to-vehicle and vehicle-to-infrastructure communication system avoids most of these issues, so it would be way more preferrable in my opinion.
The thing you fail to realize is that any self driving vehicle is absolutely going to have to possess a mobile internet connection anyway, This is not negotiable.
The reason for it is navigation. A vehicle does not just have to dodge cones on a test track. It needs to be capable of knowing its current location and navigating to a destination. Otherwise its just a cool toy with no practical utility. To do this the vehicle has to posses a map of the general area it is in.
Now, as you rightly point out this map does have to be internal. It can't just be an app like google maps because you do want the car to be capable of functioning and not turn into a brick the moment the internet cuts out or goes out of range. But at the same time this map needs to be updated regularly to account for changing conditions on the roads. Streets change names, they get closed for road works or turned one way because some official wanted to make a mark. New roads are added and old ones closed. Etc. Etc.
You don't want to end up wasting three hours driving down the highway only to discover the exit point you need to take is closed for renovation and you need to take the long way around. And as a manufacturer if that happens to a whole column of your cars it's a good way to ruin your brand.
And while gross updates and map changes can be done periodically like at night the reality is that you really, really want as much of these updates as possible to be as close to real time as possible too because road conditions change in real time and not in convenient nightly builds. Pipes burst, crashes happen, streets get closed etc. That's the sort of thing you spot on google maps on your phone or hear announced on the radio as you drive. And as a human you know how to process that and get around it. A self driving car has to have the same capability or else it is just objectively worse and people won't transition to it.
So while you will not need a constant internet connection bottom line is that if you are giving your navigation over to a computer that computer is going to have to at the very least have the capability to match the real time monitoring aspects of a driver looking things up on his phone using google maps or listening to the radio.
And the reality is that the only practical way to keep it updated is via the internet. And in particular mobile internet.
Yes, you could develop a whole new set of standards and build a whole bespoke network of communications just for these things and than deploy them all over the planet everywhere and get everyone to standardize on a single world standard. But good luck with that. People can't even standardize what side of the road you drive on. And you absolutely need standardization because you don't want cars manufactured in France to be undrivable in Germany. So any proposal that is not at least market-zone wide is a non starter.
The internet meanwhile is an already established technology that has been proven and developed. It's a standard that is well known and accepted by everyone across the planet and the infrastructure is there as well. And when it comes to internet on the go there simply is no practical alternative to mobile. Regular home wireless won't do because most people in most cities park on the street some distance away from their homes and destinations and not in convenient reach of a wireless router. And even if they were, updating at night is only fine for software patches and gross map updates but not by the minute road changes.
And while yes, coverage is a problem it is going to remain a problem no matter what technology you use. And paying to expand mobile internet is far cheaper and more efficient than paying to duplicate all the coverage for your bespoke system AND than add more on top to cover everything. So there really is no economically viable alternative.
Finally, when it comes to reliability and coverage it is simply not as much of an issue as you think because of the nature of the requirements to update.
Put simply these things are only an important issue in areas where the coverage is not overlapping and duplicated. In other words, out in really small far away places like villages of 50 people or on the open highway. But these are precisely the areas where real time updates are going to be least important. New stretches of highways opening or parts being closed off for works is something that can and is announced a long time in advance. And there are not many traffic lights in a small village of 50 people or on the open highway to keep track of. So if there are no updates in a day that's not a big deal for these zones.
And conversely in a city where traffic is constantly shifting and something is always going on and you really need those by the minute updates in order for your car that you hear on the radio or see on your satnav or phone running google maps you also to match the navigation efficiency of a human driver you also have plenty of redundant overlapping phone coverage.
Uhh, no it is not that simple. First, there will never be "an update". Any update needs to be compatible with the car software and each car manufacturer does their own thing. So you would need updates for every software variant and every version of those. Just look at how well (hint: not) Android versions get pushed to cell phones. And those have at least a common base software and a much shorter lifetime than cars.
That is not the states problem though. I mean, it's not like any country is going to suddenly create a whole new set of traffic laws that radically change how driving works to the point that you require manufacturers to rework their entire software system. As I said, most updates are going to be of the traffic information kind because that is just the baseline minimum that the vehicle needs in order to be competitive to a human driver.
And for those you don't care what the manufacturers do internally as long as you get them to agree to standardize on the same JSON packet format for the data. Something that is very easy to do.
Seconds, yes you could distribute it over the internet, maybe even a MPN of the car manufacturer (if they are still in business, that is). But for that to work, the car must be in coverage. It could be in a rural area or in undergound parking with no signal. And if the car would not start, you could not get it back into coverage.
As I explained above, that is not really an issue as the vehicle is going to update as soon as it connects to a signal and areas with little or no coverage are also those with the least population density and thus the least chance of needing rapid continuous update.
So it all boils down to this.
You have a vehicle that has to maintain an internal map of the local area anyway in order to navigate. That vehicle also has to maintain near continuous internet service to receive traffic updates and other important information in order to compete with human drivers. And the lag time tolerance on these updates is conveniently such that it favors requiring more updates precisely in areas where coverage conditions are most favorable. So at that point why not just decorate the map with additional information about traffic signs lights and have updates be streamed to the car as well?
Sure, you will want an optical backup such that the vehicle can tell if a light is red even if the signal goes out. But that is going to be a backup. Something you use out in the village of 50 people on the 3 days of the year when their only cell tower goes out. And for the city of 3 million where every centimeter is covered by 5 competing companies it just becomes a box to tick in order to make the regulators happy.