Humans do an unbelievably bad job at this, too, like all other tasks behind the wheel. Humans essentially just follow the tracks of the last car, creating ad hoc lanes. Computers can do this just as well.
On the other hand, humans also have the common sense not to (e.g.) follow the tracks of a car that has spun off the road.
I'd expect logic at that level might be more difficult to train for than the relatively low-level visual task of identifying and following tracks. But I guess existing self-driving systems already need to work at that level to predict the actions of other cars and pedestrians, so presumably the current data-driven training techniques would also handle it OK? (Assuming a lack of pranksters with shovels...)
What happen when you are the first on the road with no tracks to follow from previous cars and suddenly it turns into a blizzard/zero visibility? I am guessing the car can pull over and turn on emergency flasher...I don't know but to me I believe a Human will always have to take over at some point because you never know what mother nature will throw at you, keep in mind I am talking about a future where we aim for 100% autonomous car.
Having lived through many such blizzards, there are two things that humans do:
The smart ones pull over and turn on the emergency flasher. Then they check to make sure they have lots of gas, blankets and water, just as they prepared.
The dumb ones slow down to 45 mph and stare into the white blankness looking for clues about curves.
Exactly! Every question about how a computer-driven car will handle exceptional situations encountered by humans shares the same answer: the computer won't have got itself into the situation in the first place. Humans might drive a car into a whiteout blizzard, but that only serves to demonstrate the urgency of cutting humans out of the loop.
I'd expect logic at that level might be more difficult to train for than the relatively low-level visual task of identifying and following tracks. But I guess existing self-driving systems already need to work at that level to predict the actions of other cars and pedestrians, so presumably the current data-driven training techniques would also handle it OK? (Assuming a lack of pranksters with shovels...)