Mike Ramsey for The Wall Street Journal:
Mr. Musk has said Tesla plans to be the first company to offer customers an autopilot feature. “I think we will be the leader in autonomous cars that you can actually buy,” he said.
I still don’t know how I feel about this, but I wouldn’t be surprised to see Musk be the first to market.
He’s right that it’s a solved problem. From an algorithmic and processing standpoint, done and done. Integrating that into a usable product has been the problem so I would put my money on Tesla before anyone else.
DARPA initiated a competition for this what 10 years ago.
So all the major Universities participated in the race
which was in a desert and in a city surrounding.
You will notice if it was just money then DOD would have already
deployed it. as the code was theirs first.
all this need high sensors like HDR 180º camera, lasers, radars,
ultrasonic sensor, 3d maps, high performance GPS,
real time traffic and mapping.
redundant computer, steering and brake system.
All the algorithms being used are fake AI like neural networks,
statistics, image recognition all using high powered GPU.
Nvidia is showing 2 TFLOPS mobile GPU as solution but
probably needs more.
Volvo is already testing, Audi is going cross country.
Where is Tesla. Just PR stunt by Elon.
I call bullshit!
I think the issue is that the road tripping AI is easily confused – even if no alcohol is involved!
He (Elon) is completely wrong. Algorithms are worthless – the cars are not even as good as a good horse – when you need to get home. This is 5th century technology – or craptology – they just blind people with dazzling lights and technobabble – it is all craptology!
… maybe in 10 years there could be something that works – until then – all their blabber and banter is bunk! (in other worse – PROVE IT :P).
His words are “I view it as a solved problem.” That can be interpreted in so many ways. There are a myriad facets to this challenge. Technology is hardly the only one.
I am wondering: Would you need a special driver’s license to drive a driverless car?
Good question. Where’s the liability in case of an accident?
Malaria is a “solved problem” but it still kills hundreds of thousands of people each year (down 47% in no small part to Bill Gates). Having a solution and deploying that solution extensively enough to eliminate the problem are two different things with a huge gulf in between.
No. You are still the licensed driver of the vehicle. You are not supposed to set it and go to sleep; you’re supposed to monitor and stay alert in case of malfunction or emergency. If there is an accident, you will be held responsible as not being in proper control of the vehicle. “Driverless” does not mean out of control.
In fact, one of the major issues will be people overcompensating or taking over when the car actually can do whatever it is better. That will likely be the cause of several accidents as people adjust to the new tech.
Considering what I see on a daily basis when I’m driving around, I think the computers can already do a better job of driving than people do. My question is less about whether this works now or can work within the next 5-10 years. I think we are 90-99% there. My question is whether we are better with just a few of these robot cars out there avoiding all the jerks and idiots, or if every car should have them. As I mention in another comment here, a big problem — at least initially — will be people overcompensating for the robot and trying to correct what the robot is doing right in the first place.
“adjust to the new tech” – that’s exactly why I am asking. It’s a new technology, and it may indeed require a special permit to use it.
Seriously. I’m seeing a lot more dumbassery and obnoxious shitheaded driving out there than I remember from, say, 10–15 years ago. And it’s not state-specific, either.
I don’t doubt legislators and lawyers will attempt to say because it is new tech that we need new laws (i.e. special permits) to cover it. However, the law already says you must be in control of the car at all times. Just as setting cruise control and then not paying attention and not pressing the brake when you need to is covered, so would be turning on robot driving and then reclining back to sleep or sitting back to read. (And, yes, people will do that, and they should be prosecuted when they kill someone.)
Someone with a new driver’s learning permit is responsible if they slam into something despite the fact we know they are adjusting to new tech. So will people be responsible if they let their robot driving slam into things.
Certainly there are many bad drivers – or drivers not actually paying attention to their driving or anyone else’s – though if the driving “AI” is only better than a crappy driver – the equation is not solved.
My understanding is that in perfect conditions, systems can be setup to “drive the car”, though they are easily confused – in say rain (heavy?), snow storms, construction sites … and what happens when you leaved a nicely paved road?
What happens when the computers die, blow up or crash?
I would say a automatic driving system would allow the passengers to sleep – if they wanted – and IF there was an “issue” it would pull over the closest safe spot and wake you.
Malaria is not a solved problem. Don’t make stuff up.
Your reading comprehension isn’t very good, is it? Just because a problem has a solution doesn’t mean the problem is gone. Travel to the moon is a “solved problem” but scheduled Pan Am trips to Clavius base, or anywhere else, never happened.
Yeah, I’d interpret it as “all that’s left is engineering and grunt work” – that all the interesting theoretical problems are solved.
That doesn’t mean it could be deployed next week.
(I also don’t trust anything Musk says about Tesla and timetables, from experience with them never, ever shipping anything on time.
I’m not sure I believe the X will ever actually ship, but I’m a pessimist about this.)
Yeah, “solved problem” seems to be a term of art that confuses people who aren’t familiar with its use as one.
Malaria is a solved problem: we know how to eradicate it.
That the steps are not taken for various reasons (cost, misinformation about DDT, etc.) does not mean the solution “is not a solution” or is unknown.
It just means it’s not being implemented.
What happens when the driver has a stroke?
(Note that large aircraft are effectively flown by computers – everything’s electronic on a jet airliner, with no mechanical connection at all between cockpit and the flaps and engines and such.
Why don’t they fall out of the sky all the time because of computer failures?
Because the systems are designed for multiple redundancy and tightly written to a well-defined specification covering every failure mode anyone could think of, and tested for 100% code coverage, with parallel implementations in multiple languages. [According to someone in the industry of my acquaintance.]
(To clarify on the testing comment above for people who aren’t programmers, in this context, we mean that every line of code is run during the tests of the entire specification and error simulation suite. Any line of code NOT run is an error and removed.
And of course any output that is incorrect for the inputs is equally an error and fixed for re-test.
Thus we have a computer-run highly complex system where a REAL system failure can kill everyone on board … with a safety rate that makes cars look like the deathtraps they are.
Don’t worry about “what if the computer blows up and it crashes!!!” – that’s less likely than a wheel blowout causing a semi rig to run you over today, but you ain’t losin’ sleep over that, are you?)
That kind of automatic driving will never be allowed. (“Never” being relative to our lifetimes.) It won’t matter how good the driving computers get. Airline pilots are not allowed to sleep when they have automatic pilot turned on; neither will car drivers. (Caveat: Air laws vary by country; and generally there are crews with multiple pilots on long flights to allow for rest breaks.) Someone will have to be awake in order to take the controls in case the computer malfunctions or in case of another need. The very cases you mention (storms, construction sites) are reasons for the driver to be awake and to take over when those situations arise. In fact, in your other situations (computer crash, etc.) the AI will never meet your criteria of allowing drivers to sleep. That’s fairly unrealistic.
If the AI is better than a crappy driver, I’ll trust the AI over that crappy driver any day. I’m already considering a self-parking car for my next purchase just so I don’t have to deal with or can get assisted with parallel parking (depends on the system) — something I admit I am terrible at doing.
lol – Well “flying” a plane in near ideal conditions is child’s play. I’m sure they have real systems with redundancy that can also allow plans to land and take off – in bad conditions – though every once in a while those automated systems can contribute to problems.
Also, these flying systems are in vehicles and transport craft that cost hundreds of thousands to millions of dollars. I would guess the cost is coming down, but also, I would also guess that driving (flying at very low altitudes) is much trickier – maybe similar to landing an aircraft in a bad location, in bad weather.
I feel far safer in a car, than in any airplane. I have a fair amount of control in a car, I watch other drivers around me and i am pro active in my driving. Also I maintain my car – so I know how good (or bad) the tires are – the power (or lack there of) and the gearing. I’m sure most people do not pay attention to these things, but I watch out for bad drivers and careless drivers – which is probably 90% of the accidents out there …
Yeah, funny thing about wheels (or tires – rather) blowing out – is that most trucks (large trucks) have dualies – and at least in my state (VA) if you have retreads it can only be on dualies. So a tire blowing out (or even just a tread) will not normally cause any problems. I often see the high way littered with blown treads – I do watch tires on cars and trucks – because of that and because it tells you where the vehicle is going… but then you probably knew that already.
It would be nice to have a fully automated, tested, redundant system to drive my car – in those situations where I simply need to get from point A to B … or where maybe I’m a bit too tired or whatever (where I do not also have to be a co pilot), but that problem is not solved – maybe in 5 or 15 years.
As with many ‘tech leaders” Elon is overhyping a solution and not respecting the real problem, but then that is not new. I do respect what he has done, so far and his potential, but I I think maybe his statement was taken out of context or he is over stating things a bit.
random link
http://www.nbcnews.com/id/28678669/ns/us_news-life/t/ny-jet-crash-called-miracle-hudson/#.VQsNnCmgR3s
I suppose it would depend on the conditions – though I think the car AI should be able to drive anyone through a moderate thunderstorm, construction site, etc.. and if the system crashed a redundant system should take over – at the every least to pull over to a safe location and/or wake the driver.
I mean even back in the 80’s (or 90’s?) Audi had an unintended acceleration issue – very rare – but it was some flaw in the computer code – iirc. I would also add that computer code is (typically) much, much worse in most areas and “electronic technology/code” is often not made to fail properly – where say a building or bridge is build with lots of redundancy to prevent disasters.