It's true that we're not really there yet. And yes there are hurdles like regulation, what about heavy rain etc. But the pace at which things are going...
The autonomous super-size dump trucks drive along carefully pre-programmed routes using GPS.
We're still a long ways away from practical and safe self-driving cars. AI and computer vision just isn't there yet.
5 years is a fantasy.
Continued exponential increases in device computing power is by no means guaranteed now that chips are at the 14nm feature size.
And the recent progress in AI can only be partially attributed to improved computing power - much of it was from improved approaches to training neural networks. We could stumble into another AI winter and have to wait another 30 years for the next round of big algorithmic AI breakthroughs.
I guarantee you that self-driving cars will not be the mainstream even in 15 years. Probably more.
This time, the ongoing AI/ML renaissance is funded by private industry, and the value they are extracting from it far outstrips the costs of keeping their in-house R&D labs operating. Even if there's a contraction in the market, I can't envision any possible world where Google/Facebook/etc stop working on AI, and they have the money to keep it up as long as they want.
Want to make a Long Bet on that? I'm willing to bet up to $1000 USD, loser pays winner's choice of 501(c)(3) US charity org, that self-driving cars will be mainstream in under 10 years.
But something I can summon with an app and take a nap? 50 years may be too soon.
I don't live in the United States where every CS graduate gets a Ferrari as a signing bonus (maybe a self-driving Ferrari one day). $USD1000 is clearly a lot more money to me than it is to you.
But if you'd like to take your own side of the bet and choose your own charity, please feel free.
Moreover, I'd like to know what the terms of that bet that you'd be making with yourself. Is it for a generally available car that can drive reasonably anywhere? In San Fransisco but also in small towns without super-accurate cm-scale mapping data (which will never be both everywhere and up-to-date)? Do we consider it a success if self-driving cars blindly drive into construction sites that haven't been mapped? Or crash into newly erected barricades? Because I don't, and that's exactly where we'll be in 15 years.
Autonomous vehicles are quickly approaching what I will acknowledge is probably an 80% solution. In nice weather, when their mapping is perfect in a perfectly-mapped area, and they have a good GPS signal, autonomous cars will work well (and currently are working well in testing).
That's not the standard that's expected. No one lives in a city without road work or construction going on. No one lives in a city where temporary or make-shift signs aren't being erected, or where there are never detours, or where there isn't ad-hoc communication going on between pedestrians and drivers, or where weather doesn't block LIDAR, or, ...whatever.
We will be nowhere near level 4 autonomous vehicles in 10 years time. And not in 15 years, either.
No one in this field is confident of that anymore. Not even Google is making the same bold claims that they were 3 years ago. Their own researchers are publicly hedging now (http://gizmodo.com/how-to-teach-an-autonomous-car-to-drive-1...) and backing away from their most ambitious claims.
The only other group that I'm aware of that's still claiming that we're anywhere near having generally-useful self-driving cars are the established car makers, none of whom have traditionally been serious AI and computer vision research contributors.
"The usual touchstone, whether that which someone asserts is merely his persuasion -- or at least his subjective conviction, that is, his firm belief -- is betting. It often happens that someone propounds his views with such positive and uncompromising assurance that he seems to have entirely set aside all thought of possible error. A bet disconcerts him. Sometimes it turns out that he has a conviction which can be estimated at a value of one ducat, but not of ten. For he is very willing to venture one ducat, but when it is a question of ten he becomes aware, as he had not previously been, that it may very well be that he is in error. If, in a given case, we represent ourselves as staking the happiness of our whole life, the triumphant tone of our judgment is greatly abated; we become extremely diffident, and discover for the first time that our belief does not reach so far. Thus pragmatic belief always exists in some specific degree, which, according to differences in the interests at stake, may be large or may be small." - Immanuel Kant
http://spectrum.ieee.org/cars-that-think/transportation/self...
It (for now) requires a human driver, but operates at autonomous level 3.
EDIT: Current street view coverage: https://support.google.com/maps/answer/68384?hl=en
This is why Uber and others are scrambling to buy mapping companies. In the future the software goes hand in hand with a hyper detailed map.
They are not.
There is a reason for all of those pileup pictures whenever there is severe weather in the US.
Self-driving cars are going to be WAY better in bad weather than any human could ever be. They will be able to "see" in conditions that humans couldn't dream of. They will be able to control steering far better on slick surfaces. They will be able to respond to emergencies far faster.
I have said this before but it bears repeating, once we get self-driving cars, the insurance companies won't actually let you drive anymore.
I've driven in 42 winters on icy roads with thick snow and limited visibility, without the aid of GPS which has a funny way of disappearing in inclement weather. Google's cars have been demonstrated driving in exactly 0 winters under those conditions. Because they know it would be a disaster.
There is a reason for all of those pileup pictures whenever there is severe weather in the US.
They're driving too fast. But they knew that. They just decided to do it anyways. But they didn't drive that fast thinking that the snow was actually falling leaves or that the windshield was dirty or that the car was suddenly moving backwards, or any of the other ridiculous things that an AI will begin thinking when it falls out of a switch statement, encounters a case it wasn't trained on, or loses its GPS signal.
Self-driving cars are going to be WAY better in bad weather than any human could ever be
I have said this before but it bears repeating, once we get self-driving cars, the insurance companies won't actually let you drive anymore.
I completely agree. One day they will be completely superior to human drivers. But that isn't today and won't be for years. Many years. Like decades.
Except that autonomous cars will have lidar and radar which sees through rain and snow, accelerometers and wheel monitors to integrate position even in the face of GPS loss, and infrared and ultraviolet cameras which can see much further than human eyes.
You are right. Autonomous cars are worse right now. But they will be way better than humans really quickly.
Look at how fast the DARPA challenge went from "can't complete the course" to "had to slow the vehicles down for safety".
I'm naturally suspicious of faith-based engineering.
LOL. I like that comment. However, engineering is part science and part art. The art part often has a significant component of faith.
Moore's Law is a good example of faith-based engineering--because we had faith it would happen we committed resources to make it happen. And that operated well for almost 40 years.
I view self-driving cars in the same light--there is now faith that it will happen, so people are now committing resources to make it happen. And there don't seem to be any insurmountable problems (the biggest problems are "vision" being power and CPU intensive, but a car is a very high-density power source).
I do understand your point, though, initially many people regarded self-driving cars as "AI" and that, of course, remains elusive.
However, what the engineers quickly figured out was that you don't need to approach the problem as AI. There are much simpler heuristics--which is not surprising since driving is a background task for most humans.
Guess what? A couple years later and nav systems were better than 99% of people even in places like Pittsburgh. I haven't bought a Thomas Guide (remember those?) in 10 years, I think.
It's going to be like this for self-driving cars. It will be slow progress for a while and then it will just go exponential. Part of the slow progress is that nobody wants to have the first fatality in a self-driving car as it will be an absolute PR disaster.
The fact remains that autonomous cars are still far worse than humans in this regard.