Slashdot Mirror
Ford CEO Says the Company 'Overestimated' Self-Driving Cars (engadget.com)
Ford CEO Jim Hackett scaled back hopes about the company's plans for self-driving cars this week, admitting that the first vehicles will have limits. From a report: "We overestimated the arrival of autonomous vehicles," said Hackett, who once headed the company's autonomous vehicle division, at a Detroit Economic Club event on Tuesday. While Ford still plans on launching its self-driving car fleet in 2021, Hackett added that "its applications will be narrow, what we call geo-fenced, because the problem is so complex." Hackett's announcement comes nearly six months after its CEO of autonomous vehicles, Sherif Markaby, detailed plans for the company's self-driving car service in a Medium post. The company has invested over $4 billion in the technology's development through 2023, including over $1 billion in Argo AI, an artificial intelligence company that is creating a virtual driver system. Ford is currently testing its self-driving vehicles in Miami, Washington, D.C. and Detroit.
Overestimated the ability of the company to work on self-driving cars
Pretty sure the people creating the self driving technology are overselling its capabilities to car makers. Not to mention the real world accidents from Tesla's and the Uber incident has to place some fear of liability in car makers these days. Even with a human behind the steering wheel doesn't mean they can recover when the technology has a brain fart.
Many, including me, have said it is not going to be that easy to make a fully autimatic car. These dreams of most new sold cars being fully automatic within 5 or even 10 years etc. are not realistic.
I really don't want to be babysitting some semi-automatic car, so i won't touch the tech until it is completely and fully automatic, so i can sleep in the car while it's driving.
While I think self driving cars is in fact another level of complexity entirely, this kind of makes me think NASA/ULA vs SpaceX. How much complexity is the problem that needs to be solved, and how much is the sheer inertia of a company that has been going one direction for a very long time?
"We overestimated the arrival of autonomous vehicles"
That's fine. We didn't believe you anyway.
We overestimated the arrival of autonomous vehicles
No. You overestimated your company's ability to create a viable design within the schedule you set, and you underestimated the complexity of the problem you faced. You don't get to blame the lack of "arrival" when you're at least partly responsible for that lack.
I am so fucking tired of the weasel-worded evasions of responsibility that have become the gold standard in PR-speak. You failed to meet a date - it happens. Grow up and own the responsibility, then learn and move on. Don't make a thinly veiled claim that 'the devil made me do it' or some such rot.
'The Economy' is a giant Ponzi scheme whose most pitiable suckers are the youngest among us and the yet-unborn.
First press release: Don't buy a Tesla. We're about to release a self-driving car done right. Second press release. Don't buy a Tesla. It's not actually possible to produce a self-driving car.
Nope, no sig
Is anyone besides the CEOs themselves surprised that the corporate-types vastly underestimated the complexity of a problem they didn't truly understand in the first place?
Formula for a +5 post:
Bitcoin is the currency of the future!
Tesla makes the best vehicles and they can drive you straight to Aruba!
Everyone should have a guaranteed $100k income!
geo-fenced.. in other words it will run around a pre-defined circuit in a walled compound.
Laws are rules for the court, but merely a bottom bar to hit for life. Think beyond laws in your actions always.
I worked there in 2014-2016. I remember the CEO saying they'd have self driving cars in 5 years. Retards there generally believed that. How dumb do you have to be? Like, we're working on the tech ourselves. You can see how far away it is. But, oh no, Fields says 5 years, so it's 5 years. At least the furniture guy is realistic.
we only have computers because of space exploration!
get off this rock!
the species!
asteroid mining!
3D printing!
They couldn't even give you a car other than in black. We will let newer companies take its place.
Tesla has oversold the autopilot and people died
Uber killed someone and got the cops to try to pin part of the on the low trained safety driver that as part of there tasks is to take there eyes off the road to look at logs.
boeing 737 max 8 what happens when an sensor mess ups and that could of been much worse.
Just what is lots of cars that can fail in the same with people that have way less training as backups.
Can we finally just admit self driving cars are an Illuminati surveillance tool?
Another freedom they do not want you to have - driving.
The next step? Manual driving becomes Illegal. Don't believe me? Boy are you fucking retardedly gullible then.
Internal engineers say this is pie in the sky (2023) and it's more like 2035 for dev and 5 more years for extensive real world testing.
But that's reality.
Face it, you're more likely to have working commercial (non-military) safe fusion reactors before you see self-driving cars.
Let alone their security implications.
-- Tigger warning: This post may contain tiggers! --
In the US the most consistent roads are the interstates, but if you've driven in enough states, you know that the interstates aren't exactly uniform from place to place.
That's putting it mildly. I can tell when I cross the border from Ohio to Michigan with my eyes closed. Michigan's roads suck and are badly underfunded (only Georgia spends less per capita - they need to raise taxes but the republicans control the legislature and break out in hives when they hear the words "raise taxes".
We need to see how self-driving cars handle construction zones, rain, snow, and fog on interstates first.
I think they will figure it out but it's just going to take a lot longer than many people (including Elon Musk) are proposing that they will. I figure even best case we are at least 15 years away from a truly self driving car that could be sold to the public. And that is probably being wildly optimistic. I think the technology will make it's way in to use fairly steadily and already has but full autonomy is quite a ways off yet. I think it's a worthy goal but it's just going to take a while because it's not an easy problem to solve.
I've pretty much expected this for a while.
Don't get me wrong the progress that has been made in this field is incredibly impressive, and I have no doubt that eventually we'll be there, but I've found it laughable when you have people with kids who are 8-9 years old stating that their kids won't have to learn how to drive.
Autonomous driving is a VERY complex problem, and while it may be 90% solved, that last 10% will likely take us decades to perfect. I wouldn't expect fully autonomous cars to be the norm for probably 40-50 more years.
Heck just look at the situation Boeing is in right now. Aviation is arguably a much easier task to automate, because there are fewer other vehicles around (and the ones that do typically have transponders announcing their location), and the environment is much more structured as to procedures, yet they've had multiple planes crash due to faulty sensors and autopilot related functions.
Electric cars - sure, they'll be the norm in 10-15 years. Autonomous though? It'll be a while.
"People who think they know everything are very annoying to those of us who do."-Mark Twain
"We thought we were just going to show up at the end and sell an automated car to you rather than developing our own."
Piss off Ford.
I don't think anyone realizes how challenging this problem is.
There are practically an infinite amount of edge cases.
Invest billions in what execs think is 'Just another new product R&D cycle'
Get sold the idea that 'deep learning algorithms' are enough, just keep throwing more and more data at it!
Discover you can't get it over the finish line because NO ONE has any fucking idea how a brain 'thinks'
Legal department's analysis shows the risk/benefit ratio is so high that you'd be nuts to actually launch this hot mess
Like I've been saying all along.
The approach of the current crop of so-called, inaccurately named 'AI' is all wrong. You need a fully thinking brain behind the wheel, not just some shitty half-assed 'learning algorithm' that doesn't know the difference between a living being and a lamppost, or a real stop-sign from one painted on the back of a T-shirt, or that just because a stop-sign has some graffiti or a sticker on it doesn't mean it's not a stop-sign anymore.
If your so-called 'self driving car' needs to pull over in the middle of a trip and 'phone home' so a remote HUMAN operator can 'guide it through' whatever it is that's making it vapor-lock on you, then it's not suitable 'technology' for public roads. Period.
People like to say "everyone's stupid" and thus "if AVs are 10% better than human drivers, it will be worth it". But once you talk to the actual developers and technologists and ignore the futurists, they'll tell you that getting a machine to make decisions even half as good of humans is really, really, really difficult.
We don't give "stupid" people enough credit. The innate base intelligence of someone that doesn't regularly kill others on the road is very difficult to emulate, regardless of how we consider them in relation to the more intelligent members of the species. They don't just follow lines in the road, they adjust to lighting conditions, curvature in the road, they know how people will swerve in advance of potholes. They can quickly decide if someone/thing is about to go into the road or even make subjective judgements on how another vehicle will move on a freeway based on minute experiences in the last 30 seconds.
From a purely decision-based analysis, driving an automobile is extremely complex... still too complex for a computer to measure and judge appropriately to be autonomous. We'll get there... but it won't be quick, cheap, or easy.
And I don't believe for a second that the billion-dollar corporations investing in this technology "don't quite understand" how complex it really is. They simply put profit first, like they always do.
Tesla definitely oversold enhanced cruise control by calling it "autopilot", but far more people have died driving not-Teslas than have died driving Teslas.
Fordy co. CEO James Mackit scaled back hopes about the company's self-dancing bots this week, admitting that the first breakbots will have limits.
From a report:
“We overestimated the arrival of break-dancing bots," said Mackit, who once headed the company's dancebot division, at a Detroit Economics Club event on Tuesday. While Fordy still plans on launching its new self-dancing bot in 2021, Mackit added that "its dance selection will be narrow, what we call geo-fenced, because the problem is so complex." Mackit's announcement comes nearly six months after its CEO of autonomous dancebots, Sherry Makersmark, detailed plans for the company's new self-dancing bot in a Mediums post. The company has invested over $40 billion in the technology's development through 2023, including over $10 billion in Farrago AI, an artificial intelligence company that is creating a virtual dancer system. Fordy is currently testing its self-dancing bots in Miami, Washington, D.C. and Detroit. Fatalities have been minimal.
E Proelio Veritas.
Do they build their long term plans from what, skimming the covers of the ALWAYS overoptimistic Popular Mechanics?
No sober person who didn't have a dog in the hunt believed that autonomous vehicles were anywhere NEAR close to implementation. We still haven't solved fundamental problems with vision and processing on perfectly clear, dry days on sunny, empty California streets (where pretty nearly your dog could drive safely), to say NOTHING of the major effects of weather, night, redundancy, and the never-insignificant-in-American-contexts: LIABILITY.
If you think one company is going to seriously put truly autonomous vehicles on the road before they're essentially held-harmless if/when it runs over a kid, you don't understand corporations.
-Styopa
Also far more people have died on airplanes than have died on flying cars.
Laws are rules for the court, but merely a bottom bar to hit for life. Think beyond laws in your actions always.
Dont worry Millennials, your futures are just as rosey!
Tesla releases a quarterly safety report on their cars. About one accident for every 2.87 million miles driven in which drivers had Autopilot engaged. As a contrast, for the general public, it's about one accident per 436,000 miles.
So, much better but not perfect.
As long as it's better than driving without assistance, it's a win.
https://www.tesla.com/VehicleS...
I don't read your sig. Why are you reading mine?
Not just edge cases, absolutely stupid darwin award cases. Yes a person can physically drive down a sidewalk so you had better prepare for it.
Laws are rules for the court, but merely a bottom bar to hit for life. Think beyond laws in your actions always.
Tesla vehicles in Q4 2018 experienced accidents at a rate of 1 accident per 2.87 million miles driven with Autopilot engaged, and one accident every 1.76 million miles without Autopilot. The average car in America experiences 1 crash every 436,000 miles (these stats do not take fault into account). Effectively, a Tesla is 4 times less likely than the average car to get into an accident, and that number jumps to 6 times less likely when using Autopilot (which is constantly improving).
https://www.teslarati.com/tesl...
people die every day in car accidents from every manufacturer. take a look at the numbers next time before making baseless, misleading accusations.
i could live a little longer in this prison
Dear Mr. Hackett. I fear your automotive start-up will never succeed unless you are able to excite 'investors' into pouring money into your scheme with outlandish self driving claims. This whole truth thing will spell doom for your small firm.
Tesla has oversold the autopilot and people died
Again, nobody gives a shit about the tens of thousands of people dying from today's defective driving technology, but one person dies from an autonomous vehicle and the specific problem is fixed and suddenly the problems are intractable?
Whole planes full of people fell out of the sky in what is supposedly the safest industry because a sensor failed and Boeing is taking a few weeks to fix the problem and yet we haven't grounded all aircraft even when the sky is literally falling.
Tesla and to an even greater extent Waymo should be applauded for pushing forward with this life saving technology.
Airbags killed people too. We just have to fix the problems and move forward.
Reworked paragraph:
The final obstacles just happened to turn out to be trickier than estimated. The progress curve just started flattening out of the blue.
(A robot fouled it up, not me :-)
Table-ized A.I.
Posted as AC because I just went over my post limit, but this is basically the most hilarious post I read all day. If you take all the miles driven of everyone in the US and divide by the number of accidents taking into account unreported accidents, it comes out to around that. So what you just said is, Teslas with autopilot are as safe as any other car.
MBAs enhancing shareholder value are doing exactly what they should, that is their legal duty as office holders in a publicly traded company.
Well, good for Ford. GM and others are plunging headlong into this and creating a race of lemmings. Perhaps Ford looked at this and has decided to "pump the brakes" (haha) on the technology. It takes leadership to go against the herd and ask questions like that.
Now "leadership" doesn't necessarily equate to a correct assessment. He could be mis-evaluating the state of the technology and be flat-ass wrong. Or he could have saved the company billions of lost dollars trying to keep up with the other sheep in the herd. Only time will tell.
Yeah just ignore that the accident rate for the general public is across ALL driving conditions, while Autopilot miles are self selected as the easiest freeway type conditions in clear weather.
This kind of thing is why I lost all respect for Elon Musk, he really touts this kind of thing as if it means something.
American automakers and research institutions have been trying to get autonomous and semi-autonomous motor vehicles right for decades. Even after fairly successful tests of early autonomous vehicles in the 1990's, the technology just wasn't ready, and there was no clear path toward getting it ready. Only now are we actually beginning to make meaningful progress on that front - and I would like to emphasize, beginning.
For whatever reason people find it easy to believe that self-driving cars are either a new concept enabled by new, more powerful, more miniaturized computer electronics, or a neglected technology class that was orphaned out of a lack of perceived need. It's neither of those things. It's a brutally hard concept, exceptionally difficult to get right, which has only so far proven itself worthy in the most ideal of conditions. We're not much closer to getting it right than we were in 1997 - the cars still run into things, still stop unnecessarily, are still easily confused or tricked, and yet are still capable of driving many hundreds if not thousands of miles without incident on the open highway. They'll still plow into wildlife and pedestrians unless someone's there to man the brakes, they still don't know how to react to other drivers, they're still terrible in bad weather, and yet they can still stop themselves from getting into fender benders while pulling into or out of a parking space with a nearly perfect success rate.
It isn't as if we've never gotten them to do at least some things right, and it isn't like we haven't also gotten them to do some pretty impressive stuff - I'm actually surprised that after the early successes in the 1990's that at least some of the technology wasn't applied to semi-trucks, where it could possibly have done some good, in the same way that similar technology has already long been applied to trains. However, the same problems remain. We're still trying to teach computers, which are still barely adequate for the job, in a few short years what we've learned in over a hundred years of driving self-propelled land vehicles in both guided and unguided settings. Even with decades of automation behind them we can still barely automate trains, so what makes us think that we can get consumer automobiles to do what they can't, in a vastly more chaotic environment and without the aid of a track?
The capabilities of self-driving car technologies have always been oversold. They've been oversold for decades. They've been oversold since long before I first saw an autonomous platooning demonstration on the nightly news as a kid. They'll continue to be oversold for years to come. Eventually, somewhere, somehow, a whole bunch of someones are going to figure it out, and by then it'll be a monumental achievement in the field of robotics close to a century in the making. We've been trying to get cars to drive themselves for a long, long time, and there's still a long drive ahead of us before we get there.
Tesla releases a doctored quarterly safety report on their cars with completely made up numbers.
Fixed that for you.
Useless.
Why not compare accidents with AP enabled vs accidents with AP disabled amongst fleet of Teslas. And by accidents I mean anything that registers as accident in vehicle i.e. gets reported as such to Tesla immediately, followed by phone call from Tesla to registered owner.
Very interesting. Now to figure out their definition of "accident".
Given that their implementation of autopilot or whatever they call it nowadays is intended mostly for freeway driving and accidents mostly happen on city streets, this statistic may be useless.
They are not comparing driving under similar conditions. Tesla's autopilot or self driving is mostly used on highways.
They should compare highway accidents with AP enabled vs highway accidents with AP disabled vs highway accidents other vehicles.
If you take the automated car deaths so far and apply the ratio of regular cars on the road, adjusting for real world randomness/weather, self-driving cars wouldn't look so good.
Laws are rules for the court, but merely a bottom bar to hit for life. Think beyond laws in your actions always.
If you take the automated car deaths so far and apply the ratio of regular cars on the road, adjusting for real world randomness/weather, self-driving cars wouldn't look so good.
No fatalities for Waymo. If you compare accident rates just for Waymo it is looking pretty good so far.
The Uber fatality shouldn't count... they aren't in the same league as Waymo. Tesla wasn't fully autonomous so you are comparing apples and oranges.
Yes, it's so impressive that a car with a safety driver hasn't killed anyone.
Laws are rules for the court, but merely a bottom bar to hit for life. Think beyond laws in your actions always.
if you REALLY believe that, and weren't just trolling with an agenda, you wouldn't have ignored the fact that NON-autopilot accidents in a Tesla are still 4 times less likely than the average car. This is likely because of its advanced warning and automatic braking features -- Autopilot technology that is always active on the car, whether the driver is using autopilot or not, or even paid for the autopilot upgrade.
but haha you and i both know you're not here to have an honest discussion.
i could live a little longer in this prison
Tesla owners aren't average representatives of the population. Tesla cars are highly unlikely owned by teens or by very old people.
"It's such a fine line between stupid and clever" -- David St. Hubbins, Spinal Tap
What we really need is millions and millions of two ton autonomous networked robots rolling around our cities. What could possibly go wrong?!?!?
The difference between reusable and non-reusable rockets is physics and engineering. The difference between human drivers and self-driving cars is hundreds of millions of lines of code that all have to work perfectly. Trust me, physics is easier.
Umm, how many lines of code do you think go into a rocket launch? The space shuttle had about 2 BILLION lines of code that all had to work approximately perfectly. Any rocket launched today, especially one that plans to remain reusable, is going to have a LOT of code made to a very high standard of reliability.
And if you think physics and engineering don't play a big role in self driving cars you don't understand the problem adequately.
Now it might be fair to say that launching a rocket is easier but it certainly isn't because of the number of lines of code. The challenge with autonomous vehicles isn't doing reliable code but in figuring out exactly what the code should do. Provided a company is willing to spend the money to do it, we know how to make reliable code. (we're just not used to companies actually going to the trouble) We haven't figured out how to design software that can recognize and react sanely to all the various inputs that occur in real world driving situations. The algorithms are just super difficult to figure out. The progress that has been made has been remarkable but the corner cases still unsolved remain fairly intimidating.
So, much better but not perfect.
Not a fair comparison, people turn off autopilot when the road gets tricky. In other words, autopilot only drives the easy parts.
The average car in America experiences 1 crash every 436,000 miles
Under all driving conditions.
Tesla vehicles in Q4 2018 experienced accidents at a rate of 1 accident per 2.87 million miles driven with Autopilot engaged, and one accident every 1.76 million miles without Autopilot.
Only on freeway conditions.
Not comparable.