I know a thing or two about this technology, so for anyone curious, here's how hard it'll be to expand self-driving cards beyond the American desert.
Narrower streets: This isn't a huge issue, it should be possible to adapt whatever they have to anything with two or more lanes. There's nothing wrong with single lane roads in and of themselves, but in practice single lane roads are either alternating flow, or tiny residential streets covered in parked cars that might hide hazards, or tiny residential streets covered in pedestrians and businesses that all have to be tracked at once. All of those are a nightmare to make an autonomous vehicle deal with.
Poorly marked roads: Doing this well is way beyond what anyone knows how to do, even in principle. A very good team could figure out how to navigate a car across a random arrangement of unmarked pavement, but part of how traffic works is an ongoing social agreement inferred from the behavior of other vehicles, and as soon as you're dealing with that sort of thing, forget about it.
Wet roads: This is just a control problem, and a responsible team should be able to handle it (so, not Tesla).
Snow-covered roads. The control problem is worse than with wet roads, but more importantly this obscures road markings.
Rain: Completely destroys the quality of any sensor they're using. Probably doable for a drizzle, possibly also in a short downpour in a dry area, but the clouds of mist that form off of wet roads are way beyond what anyone is going to be able to handle for a very long time.
Snow: Not quite as bad as rain, but see snow-covered roads.
Let's not forget the classic to fuck with break assistant systems adjusted for US roads: cobblestone roads with steel tram tracks and a covering of wet autumn leaves.
I know a thing or two about this technology, so for anyone curious, here's how hard it'll be to expand self-driving cards beyond the American desert.
Narrower streets: This isn't a huge issue, it should be possible to adapt whatever they have to anything with two or more lanes. There's nothing wrong with single lane roads in and of themselves, but in practice single lane roads are either alternating flow, or tiny residential streets covered in parked cars that might hide hazards, or tiny residential streets covered in pedestrians and businesses that all have to be tracked at once. All of those are a nightmare to make an autonomous vehicle deal with.
Poorly marked roads: Doing this well is way beyond what anyone knows how to do, even in principle. A very good team could figure out how to navigate a car across a random arrangement of unmarked pavement, but part of how traffic works is an ongoing social agreement inferred from the behavior of other vehicles, and as soon as you're dealing with that sort of thing, forget about it.
Wet roads: This is just a control problem, and a responsible team should be able to handle it (so, not Tesla).
Snow-covered roads. The control problem is worse than with wet roads, but more importantly this obscures road markings.
Rain: Completely destroys the quality of any sensor they're using. Probably doable for a drizzle, possibly also in a short downpour in a dry area, but the clouds of mist that form off of wet roads are way beyond what anyone is going to be able to handle for a very long time.
Snow: Not quite as bad as rain, but see snow-covered roads.
Let's not forget the classic to fuck with break assistant systems adjusted for US roads: cobblestone roads with steel tram tracks and a covering of wet autumn leaves.