Uber's Self-Driving Cars Were Struggling Before Arizona Crash (nytimes.com)

An anonymous reader quotes a report from The New York Times: Uber's robotic vehicle project was not living up to expectations months before a self-driving car operated by the company struck and killed a woman in Tempe, Ariz. The cars were having trouble driving through construction zones and next to tall vehicles, like big rigs. And Uber's human drivers had to intervene far more frequently than the drivers of competing autonomous car projects. 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 (Warning: source may be paywalled; alternative source), 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. Yet Uber's test drivers were being asked to do more -- going on solo runs when they had worked in pairs. And there also was pressure to live up to a goal to offer a driverless car service by the end of the year and to impress top executives.

Comments

Pressured to proceed despite poor test results....

[hey!]

Why does that sound so familiar?

Oh, wait. I'm a software developer.

Self driving car hype

[Teckla]

Self driving cars are mostly hype. They're primarily self driving on very good, very clean, very well mapped roads only. Take them out of perfect conditions, and they fail miserably.

That being said, the technology is still cool, even though it has a long, long way to go. A lot of the technology could eventually be incorporated into normal everyday cars to help human drivers avoid accidents.

But the hype, at this point, is kind of out of control.

Re:Self driving car hype

[Rei]

Right. So it's perfectly okay for a "self driving car" to have a LIDAR that doesn't work, a radar that doesn't work, cameras that can't see at night and/or a neural net that doesn't work, and ultrasonic sensors don't work, and to have the "self driving car" rely on a person who's not been driving being suddenly instantly able to hop into "driving mode" during each of the once-in-every-1500-mile occurrences where the car tries to crash itself without warning. Got it! This is all totally okay.

Re:Self driving car hype

[toonces33]

A human driver would have seen her and not hit her.

Re:Self driving car hype

[aaarrrgggh]

Victim was already crossing the road, in the adjacent lane, as the car approached. The condition was caused by a bad self driving car that should apparently never been on the road.

That the root cause was someone jaywalking doesn't change the fact that jaywalking happens and is a predictable event. It also suggests that the car is inadequately prepared for avoiding hazards. (It also doesn't change the fact that when you don't have crosswalks at reasonable intervals people will improvise.)

Re:Self driving car hype

[AC-x]

And what if the obstacle was a fallen tree, or a large animal, or anything else that could be fatal to the car's occupants? I guess they shouldn't be so entitled to expect a self-driving car not to kill them either right?

Re:Self driving car hype

[fluffernutter]

It's not up to Uber or any self driving company to make stupidity a capital offense.

Re:Self driving car hype

[JonnyCalcutta]

Jaywalking doesn't even exist in the rest of the world. It is some weird concept spoken about in movies and TV shows from the US.

For example, in the UK we would call it 'crossing the road' and 'crossing the road' is not illegal (it is expected that pedestrians will use common sense). The idea that 'crossing the road' could be illegal is very strange to someone from the UK.

Re:Self driving car hype

[burtosis]

I agree. Look at how well lit the stretch of road is at night The camera view uber showed us was either purposefully darkened (just look at a histogram of a random frame or two), or the camera had an insufficient light sensitivity. The driver, had she not been texting but paying attention, would have had 5-6 seconds to do something. That said, I agree putting a human safety driver in the seat for 200 hours doing nothing except take over on a seconds notice is safety theater.

Re:Self driving car hype

[Kjella]

Jaywalking doesn't even exist in the rest of the world. It is some weird concept spoken about in movies and TV shows from the US. For example, in the UK we would call it 'crossing the road' and 'crossing the road' is not illegal (it is expected that pedestrians will use common sense). The idea that 'crossing the road' could be illegal is very strange to someone from the UK.

Germany has the same rule, I think. Came as a surprise to us from Norway, but we crossed so far from traffic nobody fined us or anything. But then Germany is notorious for having rules for everything and actually sticking to them. It's kinda nice and incredibly frustrating at the same time, depending on what side of the stick you're on.

Re:Self driving car hype

[fluffernutter]

I think that's obvious, but it doesn't change the fact that self driving should be held to a standard where they can accommodate people doing stupid things, otherwise the whole line about saving lines is just blowing smoke. Defensive driving is a 'thing' for good reason and very little of it changes if a computer is driving. If you see something curious, you slow down. Simple?

Re:Self driving car hype

[goose-incarnated]

And yet if a human had hit her they wouldn't have been judged to be at fault.

Yes, they would. Drivers are routinely held to be responsible for hitting almost stationary objects in the middle of the road. That is irrelevant anyway. What's relevant is that the car did not see an almost stationary object in the middle of the road. That the object happened to be an old lady pushing a bike is irrelevant.

Re:Self driving car hype

[Knuckles]

Germany has the same rule, I think. Came as a surprise to us from Norway, but we crossed so far from traffic nobody fined us or anything. But then Germany is notorious for having rules for everything and actually sticking to them. It's kinda nice and incredibly frustrating at the same time, depending on what side of the stick you're on.

In Germany it depends, but is is not illegal to cross the road. The law is:

(3) Wer zu Fuß geht, hat Fahrbahnen unter Beachtung des Fahrzeugverkehrs zügig auf dem kürzesten Weg quer zur Fahrtrichtung zu überschreiten. Wenn die Verkehrsdichte, Fahrgeschwindigkeit, Sichtverhältnisse oder der Verkehrsablauf es erfordern, ist eine Fahrbahn nur an Kreuzungen oder Einmündungen, an Lichtzeichenanlagen innerhalb von Markierungen, an Fußgängerquerungshilfen oder auf Fußgängerüberwegen (Zeichen 293) zu überschreiten. Wird die Fahrbahn an Kreuzungen oder Einmündungen überschritten, sind dort vorhandene Fußgängerüberwege oder Markierungen an Lichtzeichenanlagen stets zu benutzen.

-- https://www.gesetze-im-interne...

Translation:
Someone walking on foot has to cross roadways, while heeding vehicle traffic, speedily on the shortest path perpendicular to driving direction. If the density of traffic, speed of traffic, visibility conditions, or the flow of traffic require it, a roadway must only be crossed at road intersections, at traffic lights, on the inside of markings, or at pedestrian crosswalks. If crossing the roadway at road intersections, any available pedestrian crossings or markings at traffic lights must be used.

I.e., you cannot simply cross just anywhere on a high-level road with dense, fast traffic (think Autobahn) or in really bad visibility like dense fog. But a road like in the accident video, even if it's dark, is just fine if there is no dedicated crossing nearby

Re:Self driving car hype

[Knuckles]

What's relevant is that the car did not see an almost stationary object in the middle of the road. That the object happened to be an old lady pushing a bike is irrelevant.

Exactly. And any modern non-autonomous car with "simple" collision avoidance system would have noticed. Dunno what Uber is doing, but it does not come as a surprise that it's Uber who are the most irresponsible and reckless

Re:Pressured to proceed despite poor test results.

[Anonymous Coward]

What I'm shocked about is that, of all companies, Uber would make morally dubious decisions in its race to profit off of a new market. I mean, when have they ever acted like that before?

You want good? Or cheap?

[BeerCat]

Clearly, not all autonomous vehicles are the same.

It's very like camera - a cheap one and an expensive one will both offer "autofocus" and "zoom lens"

The cheap one will have 3 or 4 focus settings, while the expensive one will be continuous. The cheap one will have 2 or 3 zoom settings, while the expensive one will, again, be continuous.

So, Uber's cars look to be at the "what is the minimum that can make a car steer itself" end of the scale, and the Google ones are "have we missed anything off the long list of things that will help a car steer itself" end

Not just a jaywalker....

[Anonymous Coward]

Agreed. Also, how many of us have had kids run out in front of us? I have. And if I weren't paying attention, the kid, his parents and I would have a very bad year. Who cares who fault it is: I was part of hurting someone.

And so what if someone was an "airhead" and stepped out in front of the car? Doesn't make it OK. The purpose of self-driving cars is to make the roads safer because the machines are supposed to be better than humans.

Re: Uber hatred turned political a long time ago

[Anonymous Coward]

That's a baldface fucking lie, and you know it. Uber has come under massive criticism from day 1 for shamelessly and egregiously breaking livery and employment laws in nearly every jurisdiction in which they have established themselves.

Now their half-baked AI implementation has killed someone, and you want to beg off criticism of blatant criminality as merely political grandstanding? These mobsters deserve every ounce of criticism they get regardless of who is in office.

Re:That's exactly what the parent said.

[AC-x]

If that machine is incapable of detecting and avoiding unexpected obstacles and that unexpected obstacles is less squishy than a human it could easily be the occupant of said machine that dies...

Re:Pressured to proceed despite poor test results.

[complete loony]

Sounds like just about every failed IT project. Rush to market, ignore test failures, probably a thermocline of truth.

From what we've been hearing, somebody in the chain of command between the inattentive driver and the CEO, deliberately created this situation and should be charged with manslaughter.

Re:Uber hatred turned political a long time ago

[stabiesoft]

Did you not even bother to read TFS? In the summary it quite clearly states the other self drive companies were achieving 5600 miles between interventions while uber could not meet it's 13 mile goal. Sounds to me like uber's system is just plain not ready. I'd even question 5600 miles. Once they get to 1 million I'd say they are there.

I think you are the one making this political.

Why set the bar so high

[monkeyxpress]

Self driving cars are mostly hype. They're primarily self driving on very good, very clean, very well mapped roads only. Take them out of perfect conditions, and they fail miserably.

But even a car that could drive under such conditions would be extremely useful. Take, for example, public buses. They just drive around the exact same route everyday, and the route in many places is upgraded with special lanes and signalling infrastructure to make their job easier. There is no reason why you couldn't start with replacing such bus routes in cities with moderate weather conditions. Over time a combination of roading infrastructure improvements (special lanes, intersection redesigns, beacons etc) and the tech getting better could easily expand out to cover the majority of vehicle uses in a city. Again, we do this for bicycles and buses, so why is it impossible to imagine it would make sense to do some road works to cater towards driver less cars?

Another example is motorway driving. Motorways are already an extremely controlled and regular environment. It would be great to have a driver less truck that can go door to door, but there is no reason why we can't start with depots built off the side of motorways where local human drivers pick up and drop off longhauled trailers. As for weather conditions guides in the road way and other navigation infrastructure could be added if these problems cannot be dealt with using lidar and cameras.

Yes, I agree that a car you can just dump into an unknown urban environment is a long long way off. But I don't understand the fascination with meeting this goal before driver less vehicles can be useful to us.

Re: Pressured to proceed despite poor test results

[hey!]

Someone is dead because of a faulty development process, which in turn is the result of a toxic business climate.

I suspect this happens more often than we know; it's just seldom that you can connect the dots so readily.