I have spent the last 18 years working in autonomous vehicles, IE I was riding in a autonomous truck in 2001 taking data, and looking through it, and have done this in a dozen places in 3 different countries. Trust me you way over estimate what the state of the art is actually doing. Tesla's accidents were not just a over site, they are a current issue with the state of the art.
> Cars already have temperature sensors,
> cars already have skid sensors.
Think about that from the computers position. Ice, snow do not melt instantly. It can alert on temperature, and it can alert after you have slipped. Not before. People use tons of clues computers just cannot grasp today. You will smell the rain, you will have driven on this same surface, and will pull in that background to sense a subtle color change, IE pick the sun position, the sign position, take the shadow positions into account. For example it could easily be well above freezing in the morning, the road could be clear everywhere but in a corner that is shaded from the sun, autonomy enters that corner at 75 mph, damn right that car is going to know moments before it crashes, that it is now on ice.
To make a safe autonomous car that is not way too often a slow moving hazard, it has to have outside help. That same corner in the exact same conditions the last 10 times you went through was fine at 75, but because it didn't rain in freezing temps the night before, and when it did was on a weekday, a hundred cars had kept the tracks clear with the heat of tires. But this is Sunday, not enough traffic to clear the road from that midnight rain...
Very similar with the stupid slick warning from cars, they are wrong most of the time. But who cares, the driver decides if it applies this time.
> Advanced systems that rely on a stereo pair of cameras (e.g.: Mercedes, Mazda, etc.) and/or a LIDAR complementing the video feeds.
I have worked with those sensors, the LIDAR is insanely expensive to do and parse, and live on a rough road, and mount, also it is only necessary in a few situations. The Radar cannot pickup minute movements and distinguish a statue, from a horse, or a light pole from a person even with a stereo camera, until they move significantly. The car is moving, then angles and distances and timing are not precise enough due to environmental factors, are highly color and background noise sensitive... For this reason they will track things outside your path, but do not react to them.
>Outside of stationary sensors streets, the car will be unable to drive autonomously.
No, You don't need the sensors everywhere,
>The best would be for cars to optionally benefit from extra information.
Correct, that is 100% what I was saying, you need them where auto cars have issues. If the external sensors break or are missing, the car slows to 40 over that hill it cannot see over. It requires the driver to put their hands on the wheel, and refuses to take that path without a driver. Or it will only follow at speed once a trusted communication from another equipped vehicle has taken that corner and verified it...
Basically I could see a future where 10% of the cars have precise sensors, those can go more places, and can lead, have cheaper insurance. Perhaps they get paid for their service. Not needed in enhanced areas of the city, and on interstates. The cheaply sensor-ed cars need to always have human assistance to operate outside these covered areas. A cheaply equipped Uber auto car, may have to schedule a precisely equipped escort through any tough areas, or a drone fly over. Taking the auto car to a city without augmentation may cost much more, costing the businesses in those areas, so they will move, or pay for them...