I think it will be easy to get to the 80-90% but the last bits are going to kill this. Bad weather conditions, night, extreme situations, etc. Self driving cars have to drive better than the median, because the median includes drunk and elderly drivers dragging it down. I'm sure a self driving car will drive better than a drunk but that's not the goalpost, the goalpost is you and me, intelligent rational drivers who basically never have accidents (that are our fault). This is going to be very difficult as the situations that arise often times require an intelligent decision, intelligent prediction. I just don't see all those machine learning dumb systems getting there, not removing the wheel. And until the wheel is removed its useless. A car that drives but requires me to be aware enough I can take over on a split second is not self driving, so the bar is very high.
The impression I got from your statement was that you consider fully autonomous vehicles not far off. I disagree.
The point I was making was that the vehicles you see driving around are literally fair-weather vehicles.
A vehicle which can navigate autonomously 95% of the time might not be that far off (I still call it a decade though), but those last 5% (thus, no steering wheel at all) are going to be way harder. To give but a few examples: humans are taxed by navigating roads in snowy weather at night, but they usually pull it off. Or intersections controlled by traffic police. Or intersections with undisciplined drivers pushing in. I'll be interested to see an autonomous system deal with such situations gracefully.
I agree that there are difficult regulatory issues to figure out, but don't underestimate the technical hurdles.
And my point still stands. How do you define a "fully autonomous vehicle" is an important question. Consider the following three possibilities:
- Drives better than the median human driver in average weather conditions on half of the roads in the US (alternatively, for harder goals, replace with US+Europe or World) - Drives better than 95% of people in 95% of weather conditions on 95% of roads in the US - Drives better than any human in any weather conditions on any road in the US
I'd argue we're almost certainly past the first milestone, and probably at or approaching the second, and the third will never be reached, but that doesn't matter.
>A vehicle which can navigate autonomously 95% of the time might not be that far off (I still call it a decade though), but those last 5% (thus, no steering wheel at all) are going to be way harder. To give but a few examples: humans are taxed by navigating roads in snowy weather at night, but they usually pull it off. Or intersections controlled by traffic police. Or intersections with undisciplined drivers pushing in. I'll be interested to see an autonomous system deal with such situations gracefully.
Fwiw, I'm fairly confident that waymo's vehicles can handle flagmen and have more experience driving in the snow than I do. And I'm sure they can handle aggressive or undisciplined drivers.
Conversely, there are situations that AVs will handle much much better than humans already (a neighborhood with a lot of kids running around and shrubs occluding them for example).
For context, we know that the first goal (better than average driver in half of road conditions) is now achievable by teams of 1-3 people on a tight budget in O(months) (and really, maybe weeks), since people have done that and posted it on the internet.
> I'd argue we're almost certainly past the first milestone, and probably at or approaching the second, and the third will never be reached, but that doesn't matter.
I agree on the first and third milestones. Or kinda -- perhaps the first milestone is only reached for California, or for US highways yet.
I think you underestimate how difficult the second milestone is. The variation is tremendous, even merely within the US, to say nothing about the rest of the world.
> Fwiw, I'm fairly confident that waymo's vehicles can handle flagmen and
I'm sceptical. I think there will be so much ambiguity, both through road situation and through human interaction which we interpret with comparable ease, that the car may be stumped. And we set the high hurdle of forgoing steering wheels, so we need some other, yet to be devised way to give the car hints safely and efficiently.
> have more experience driving in the snow than I do.
I wasn't thinking about handling the car (I can easily believe that a computer does a better job of that), but rather about decreased visibility, no visible road markings, etc.
> And I'm sure they can handle aggressive or
Fairly easy I guess. Just be defensive, put a bit of distance between you and them.
> undisciplined drivers.
Much harder. Not from a safety perspective (again, just be defensive), but from a I'd-like-to-arrive-at-some-point perspective. Because then again it becomes about human interaction (being assertive without being an asshole, muddling your way through this blocked intersection through essentially non-verbal communication with the drivers around you), and computers will (inherently, perhaps) be worse at that than humans.
> Conversely, there are situations that AVs will handle much much better than humans already (a neighborhood with a lot of kids running around and shrubs occluding them for example).
Except, again, that humans will be able to judge body language or the rules of a game the children are playing, and anticipate instead of being confined to reacting (which the computer will indeed be better at, I agree).
Perhaps my pessimism also stems from the fact that the oh-so-smart driving assistants in my car are so shitty (read my other posts if you're interested) :-)
EDIT: Thanks for the interesting discussion so far.
My anecdotal experience shows that there exist vehicles capable of tooling around on highways and city streets without the need for human interaction.
>but from a I'd-like-to-arrive-at-some-point perspective. Because then again it becomes about human interaction (being assertive without being an asshole, muddling your way through this blocked intersection through essentially non-verbal communication with the drivers around you), and computers will (inherently, perhaps) be worse at that than humans.
There's a good video I've seen about how the Google cars can increase their assertiveness until they can go.
>Except, again, that humans will be able to judge body language or the rules of a game the children are playing, and anticipate instead of being confined to reacting
Not if they can't see the children because, as I said, they're behind trees.
>Perhaps my pessimism also stems from the fact that the oh-so-smart driving assistants in my car are so shitty
Indeed, that's why I clarify that Adaptive Cruise Control isn't the same thing as an autonomous vehicle and you shouldn't use one to judge the other.
That's one of the things that most strikes me about driving in Manhattan. In large areas of Manhattan, at least at certain times of day, you have to be almost insanely aggressive from the perspective of tooling around Mountain View on the average day. Otherwise, you're not going anywhere and are going to have everyone blaring their horns at you. (Plus, pretty much every pedestrian will take advantage of your timid driving to cross the street in front of you while you're hesitating.)
You got what I was aiming for, yes.
That plus difficulty in understanding streets obscured by a cover of snow, etc.
> Less prone to be overconfident in their stopping/steering abilities.
Very true. Computers have weaknesses, but vanity is not one of them :-)
> I'd probably trust a computer in the snow vs. most humans.
I have a car (2016 model) with several fancy assistants.
The lane-following assistant (which only works in very limited, benign conditions anyway) is regularly confused by shadows on pleasant days. Or by puddles on less pleasant days.
The car-following cruise control has missed a 40t semitrailer and would have merrily accelerated right into it.
The car-following cruise control regularly panics because it suddenly interprets guardrails or street signs as cars. On an empty road. This is not nice, because it has braking authority.
The computer in my car, at least, can absolutely not be trusted.
Even if it can only navigate between SFO and Mountain View, that's ready to deploy as a commercial venture. There will be an adoption ramp anyway. It will just follow the driving difficulty gradient.
In practical terms, it's ready now. Driving people home from the ski area in a blizzard will be the final task, but it doesn't need to be solved to put both Uber and GM out of business.
I (and others, I believe) were arguing when a car we would want to buy could forgo a steering wheel.
You're right that e.g. Uber could have a steering-wheel-less fleet that they simply retire in adverse conditions, sooner than that.
I wasn't aware I did that.
> but those sorts of cars are imminent for at least early adopters
without steering wheels? Dear God I hope not.
I use driving assistants in my car, but I'm painfully aware of their limitations. These days we absolutely still need a steering wheel. In my car for mundane situations, in Elon's miracle machines maybe only to park in an utterly unmarked field at a festival or whatever.
I dunno, maybe some kind of wheel in front of the human, which they could turn left or right to command the car to change direction ;-P