Please consider Uber’s Self-Driving Cars Were Struggling Before Arizona Crash.
Waymo, formerly the self-driving car project of Google, said that in tests on roads in California last year, its cars went an average of nearly 5,600 miles before the driver had to take control from the computer to steer out of trouble. As of March, Uber was struggling to meet its target of 13 miles per “intervention” in Arizona, according to 100 pages of company documents obtained by The New York Times and two people familiar with the company’s operations in the Phoenix area but not permitted to speak publicly about it.
Not all drivers followed Uber’s training. One was fired after falling asleep at the wheel and being spotted by a colleague. Another was spotted air drumming as the autonomous car passed through an intersection, according to the two people familiar with Uber’s operations.
A video shot from the vehicle’s dashboard camera showed the safety driver looking down, away from the road. It also appeared that the driver’s hands were not hovering above the steering wheel, which is what drivers are instructed to do so they can quickly retake control of the car.
Hand Above the Steering Wheel?
Try it yourself, right now. Stretch your hands out over your computer, grabbing nothing and see how long you can hold the position.
This is not a case of the technology not being ready, this is a case that Uber’s technology sucks.
Waymo is ready. Uber isn’t.
Mike “Mish” Shedlock
Wrong again! Lots of people try to be truck drivers but because of the poor pay and poor working conditions, the turnover rate is very high (something like 95%). This has happened not just in the US, but also in that other victim of Reaganomics/Thatcherism, the UK – link to independent.co.uk
A page from Paccar’s (think Peterbuilt, Kenworth, DAF) shareholder information this year:
link to s3-us-west-2.amazonaws.com
As far as testing goes, deciding on test venues based on political whim, rather than simply letting each punter test where he feels it most appropriate, is certainly suboptimal. In our financialized dystopia, cities hooked on Fed freshprint driven asset appreciation, fall over themselves to accommodate AV testing.
Not because their bicycle riders particularly want half-baked AVs making their already unpredictable traffic more so; but instead because those donating to their politicians want as much freshprint pumping up their “assets” as possible.
So, you get testing in Phoenix, rather than in Tonopah and Gerlach. On suburban streets, rather than on Nevada freeways. With somewhat predictable consequences wrt the risk of hitting the occasional pedestrian.
Congrats – Now learn to think. Generally, it’s more important.
Mish said: “Thanks. Help in combatting stupidy is always appreciated. Look in a mirror any time. BTW, revoking UBER was fully warranted.” At least I know how to spell.
Thanks. Help in combatting stupidy is always appreciated. Look in a mirror any time. BTW, revoking UBER was fully warranted.
Mish quote: “If you cannot test, then you cannot have the technology because idiots will preach “it’s never been tested”. I really wish people would think before posting stupidity. Alas, I ask too much. We had controlled testing, with a driver behind the wheel. Is there any other way to test? The technology failed but the person J-Walking at night, in dark clothes gets about 90% of the blame. People do stupid stuff, get hurt, idiots scream.”
Don’t worry MISH, I will let the the Governor of Arizona know just how “stupid” he is for taking these cars off of the road.
Re self driving trucks: link to technologyreview.com
@Kinuachdrach – “why the economically irrational…”
Because cars are cheaper and easier to work with than big rigs. Because the market for cars is far larger. Because the environmental and use-case variety is far larger for small vehicles than big rigs.
The idea of a back up human driver is destined to fail for the same reason that management by committee or socialism does. If you believe you aren’t the one primarily responsible for results you stop caring or, in this case, paying attention.
Mish: “There is no purposeful bias against truck drivers. There is a shortage of them, as no one wants the job.” Which would seem to reinforce the hypothesis that the fastest economic return from development of Autonomous Vehicles would come from the long-distance trucking sector. So why the economically irrational focus on the automobile sector? My old economics professor used to say that, whenever we see something happening that apparently does not make sense, there is a government regulation somewhere.
“There should be no difference between night and day driving for an inanimate machine. One might reasonably have believed that surrounding the vehicle in a cloak of technology would have been sufficient to save the pedestrian. But evidently not.”
Of course, there is a difference between night and day. It’s ridiculous to state otherwise. Expecting the technology to work at night is another matter. I agree it should have worked, at least better than a human. It failed. But probably no worse than a human. Over time – quickly – technology will be far better than humans.
Don’t fall for media campaign which first shows video of pedestrian popping up in cone of light in front of car, then says obviously a human couldn’t avoid the collision either. This is not a legitimate comparison. Unlike the new wave of vehicles, a human actually is intelligent, and knows there is no point in looking into the dark while driving at night; instead, people focus biological vision on the cone of headlights. And no surprise — reaction time is short and insufficient to save the pedestrian. However, a self-driving vehicle does not “see”. It is not focused on the cone of light; rather a suite of sensors using radar-like and sonar-like technologies should be scanning all around the vehicle to keep it aware of objects in order to compute potential collisions. There should be no difference between night and day driving for an inanimate machine. One might reasonably have believed that surrounding the vehicle in a cloak of technology would have been sufficient to save the pedestrian. But evidently not.
I don’t know that it is inevitable. Most technology fails. The problem with AVs is that they have to take every edge case into consideration. And nobody knows what all of those actually are. I think that is a much bigger deal than most optimists consider. For this technology to be accepted, there also has to be a big enough market for it to make it more profitable than existing systems. I think there might be value in replacing truck drivers and cabbies, but is that a big enough market? Given all the hype and expenditures? The best article I’ve seen on the subject discussed that it will never work, in mass, without substantial modifications to our transportation infrastructure to make it less human-friendly and more machine-friendly. But that would take massive government expenditures, which I’m not sure folks in this country are particularly keen on.
“Dare we say it might have something to do with cultural bias against working class truck drivers?” – How about 85% of the people of age drive cars. How many drive trucks? There is no purposeful bias against truck drivers. There is a shortage of them, as no one wants the job.
There does not appear to be much economic advantage to the autonomous commuter private car. So why are the Big Guys focusing on this, rather than on trucking? Dare we say it might have something to do with cultural bias against working class truck drivers?
If a technology is going to be adopted, it has to offer an economic advantage. It seems like the low-hanging fruit for autonomous driving is long-distance trucking, especially over freeways. Then depots on the edge of conurbations where a human driver would take over for the tough part. This could provide material efficiencies in terms of both equipment and driver usage — at the expense of motel & restaurant operators, since there are usually both winners & losers in any advance.
You’re right, but that’s one of the reasons I think it will take longer than the optimists believe. Each mishap will be headlines.
A drunk or texting driver would have killed her just the same. This happens every freaking day. Human drivers kill about 100 people in the US each day, but most of us are desensitized to it and don’t pay attention. If robot cars can get that statistic down to 10 per day it will be a vast improvement, but because of the focus and attention there are going to be furious headlines about every robot car accident for many years.
Notice children the dent on the passenger side, made as the bicyclist was moving from left to right. Even the least bit of brake reaction might have saved that persons life
Airline safety is a whole lot better than it used to be or so it seems. They’ll get it right but all learning curves are this way, look how long it took society to control its nuclear weapons?
The whole thing sound Orwellian to me, but I think the transition is going to be tougher than the optimists believe. It might take a government mandate at some point for all new vehicles (which could create bull market in used cars.)
And the flying car and jetpack people.
And the flying car and jet pack people.
Nearly every morning on my commute to work I hit a traffic light coming out a parking lot of a Costco & movie theater, even though it is 7am and neither will be open for hours. Yet, I am supposed to trust driver less vehicles?
What’s a few human victims in the pursuit of progress? We don’t even have smart traffic lights and we’re converting to driver less cars in 10 year years or less? I don’t see it.
Customer beta testing is where the problems are ferreted out. How about each new software revision gets named after the dead person that inspired it. “ALERT- Download software upate Uber Elaine 1.0 as soon as possible”
What is the purpose of driverless cars if you have to drive them?
Did not know Uber had these data. Assuming the measurement systems and criteria are similar, one of these systems is obviously all over the floor. It seems to me the question now is: Was it profound ignorance of how to go about shaking down a complex system or just old fashioned criminal negligence that killed this poor lady?
(when I hit return it enters the message, so it is split) And we are putting more bad technology. Cruise Control is probably fine, but lane departure, collision avoidance etc. where you will pay less and less attention. And you can’t reacquire what’s happening on the road in a fraction of a second. That is a worse problem than keeping your hands on the wheel.
This is something Humans are bad at. The driver fully attuned to the road going beyond the speed limit but paying total attention to the road is safer than the one barely looking and relying on the technology.
It is more likely to be like the microwave oven. The microwave oven was thought to be something that would displace conventional ovens and revolutionize cooking. The microwave oven did go somewhere though. It is now used to cook TV dinners and to reheat leftovers.
“it may never go anywhere”
the horse and buggy people likely thought the same thing
At 5600 miles between interventions, Waymo is nowhere near “ready”. IIRC, Waymo was at a little less than 600 miles in 2015 and then jumped to 5200ish in 2016. I thought: “wow, at that rate of increase, they’ll be ready in a few years.” But then they only jumped a few hundred miles in 2017. I don’t know if that means the easy fixes are done and the next stage is intensely difficult, or they’re just taking a break to tweak some issues. But they’ve definitely gone from exponential improvements to logarithmic. And Waymo vehicles are still only driving at 30 mph. In my mind, we’re still at the very early beginnings of this technology. It may never go anywhere.
Ain’t technology grand? Like any war, there’s always going to be casualties. Just hope that you’re not one of them — like that guy that was incinerated in one of Elon Musk’s Tesla cars last Friday in California. C’est la vie (That’s life!)
“Waymo is ready. Uber isn’t.”
….and all the socialist paradises that hasn’t collapsed yet, are really wonderful utopias…
If Waymo isn’t crashing, the main reason is they are more carefully limiting the complexity of the situations they expose their cars to. While Uber attempts to confront the real world more prematurely.
I took a drive past the fatal accident site about an hour ago (about 20 min from where I live). There is only one route to the site and requires going west on Rio Salado through the university night life district and making a turn on Mill Ave. College kids are all over the place so you have to be careful. Once you cross the town lake, you pass under route 202 and suddenly pass the spot before you can snap your camera. Not that it matters because the Marquee Theater has all of its lights on now so the place is well lit. The map on the twitter page is upside down if you consider north to be ‘up’ as most people do. In any case, with the theater lights on – possibly as a result of the fatality – there are not likely to be any accidents there in the near future.
Keeping your hands on the top of the steering wheel causes anatomical problems over long periods of time. When I’m driving across the country, I keep them low so they can rest on my lap or knee.
“Waymo is ready. Uber isn’t.”
And what data do you have for arriving at that conclusion? On the one hand, Waymo’s press release making unverifiable claims. And on the other hand, Uber’s internal company documents obtained by the press with help from two anonymous insiders.
Not exactly apples vs apples.
Security guards fall asleep in their boots waiting for something to happen. Same, same. How long can you sit and watch a clock?
—–>My question is what type of driving was that 5,600 miles? Are we talking some completely staged place like a lonely parking lot going around cones? Or are we talking the car is driving itself entirely though complex city streets in say LA or SF, stopping at 4 way stops, stopping for school kids radomly jaywalking, swerving out of the way of a piece of junk in the road, dealing with broken flashing stoplights,…..etc….
“tests on roads in California last year, its cars went an average of nearly 5,600 miles before the driver had to take control from the computer to steer out of trouble”
One must calibrate miles-per-intervention numbers. Such numbers want to be fairly low. If they are too high, the environment is ‘solved’. Time to move development on to a harder-to-drive environment. The question is, What is the best ‘fairly low’ number? Each of the teams answers that question as best they can.
This accident might kill Uber’s self-driving team. It will be interesting to see what behavior we see from the next team in this field that has a disaster.
And software guys all over the world are examining themselves, wondering what they would have done if the “bug” had been their’s. Life-and-death stuff is normally pretty hidden from view and committee-ized. This one comes from real life everyone can empathize with.