Uber to begin testing self-driving cars in Pittsburgh

The ride-sharing company wants to ditch the human element.

Read the whole story

 

[tetrapyloctomy]

If they don't slow down in the Squirrel Hill Tunnel for no other reason thsn, "Ah! Tunnel!" then they're already better than the average Yinzer commuter.

 

[DOOManiac]

One step closer to Johnny Cab!

 

[spoof]

2030 eh? Maybe someone will fix the roads by then. A good car, driver or not, needs a good road.

 

[nehinks]

You know, " almost every expert Ars has spoken to thinks 2030 might be a realistic target" can just as easily mean they all think there's maybe a 10% chance of hitting that date. The phrasing can be taken decidedly ambiguously.

 

[DanNeely]

Well that's one way to solve the problem of labor disputes with drivers.

 

[Statistical]

I understand why Uber wants to go driverless but it has to be pretty shitty for Uber drivers when the profits they make for Uber are used to research ways to get rid of them.

 

[Statistical]

Well that's one way to solve the problem of labor disputes with drivers.

Well at least until the Artificial Intelligence Citizenship Act of 2087.

 

[DanNeely]

Jamie4443[/url]":wcdnszw8]Pittsburgh is a notoriously corrupt city, along with the whole state of Pennsylvania. There are laws in "the 'burgh" that prevent you from claiming damages in a lawsuit from emotional distress, for example. Part of the cover-up of the child sexual abuse at Penn State was because of the state's special secrecy laws that apply to universities. And a lawyer told me that car crash lawsuits in Allegheny County (Pittsburgh) have much lower awards and higher dismissal rates than the national average. And there's probably a lot more.

I wonder if Uber has chosen Pittsburgh for these reasons, having already assessed the legal environment there and possibly even "primed" the legal system in advance. Not to mention sizing up the local population that is, shall we say, a bit backward, impoverished, and as they say locally, "ignurnt." The poster above refers to the locals as "Yinzers" because they say "younz" or "yinz", meaning 'you-uns' or 'y'all' for the plural of 'you'. It's quite a jarring Appalachian dialect that's hard to understand. (For the record, I lived there for 17 years).

Driverless cars are crashes and liability lawsuits waiting to happen and I'm amazed that they're taken seriously anywhere. But if they happen in Pittsburgh then Uber will likely get away with it.

Uber came to Pittsburgh because of CMU's status as a major robotics research center, including for autonomous vehicles. There was a major group working on self driving mine vehicles for Caterpillar (IIRC a joint venture with CMU); that gave them a large established team to hire away en mass. (Source a former co-worker of mine who left for Cat about a year ago.)

 

[LordDaMan]

They picked Pittsburgh because the road system is *horrible* here. Streets all willy nilly with no central planning, pot holes, really bad drivers, etc. I know, i drive it every day

If it can drive here and not get into an accident, it can drive anywhere

 

[SLee]

DanNeely[/url]":2qgln1es]Well that's one way to solve the problem of labor disputes with drivers.

We'll still need to see the economics of it, drivers as it is are cheap and (as noted by the article) provide the capital (vehicles) as well as maintenance. With SDCs, that all falls on Uber. And drivers can be used to influence local politicians (job creation, feed my children, etc); you can't do that with robot cars.

 

[tenaku2]

I'm intrigued that they are doing autonomous driving in a city with real weather. Almost everyone else is testing in the southwest, which is about as close to 'controlled conditions' as you can get, 90% of the time.

 

[Statistical]

DanNeely[/url]":2s1lg231]Well that's one way to solve the problem of labor disputes with drivers.

We'll still need to see the economics of it, drivers as it is are cheap and provide the capital (vehicles) as well as maintenance.

Uber is already indirectly paying for that right now (in the form of driver compensation). Well if they aren't then collectively Uber drivers are working to lose money. Now that may be the case but if so then their model isn't sustainable.

Fee paid by Uber to drivers = driver's capital cost + finance cost + maintenance + fuel + driver compensation.

 

[JButler]

...although there will be a human operator in the driver's seat at all times to take over at a moment's notice.

Right. And it will be like this basically forever until we overhaul the road infrastructure completely.

almost every expert Ars has spoken to thinks 2030 might be a realistic target

Ha, that sounds much more realistic than Google and its followers here.

Problem is, I don't see what they can do in 2030 differently than what they do now. Sure, CPUs will be faster, sensors will be more accurate and cheaper. But they still won't let you overcome the crucial software modeling, complexity and reliability issues they face today and will likely increase over time. The only way out is to drastically simplify the driving condition computers have to deal with. But then it becomes a chicken and egg problem because we can't just lay down a massive Disneyland monorail-type infrastructure everywhere.

 

[lewax00]

I understand why Uber wants to go driverless but it has to be pretty shitty for Uber drivers when the profits they make for Uber are used to research ways to get rid of them.

It's hardly unique though. E.g. a factory uses its revenue to hire engineers to build machines to automate the jobs of assembly line workers.

That's just business. Workers make the company money, the company uses that t money to find ways to make more money, and one of the major ways to do that is cut costs, and workers are expensive.

 

[Jeff3F]

I think the 2030 estimate is too far out, machine intelligence capability is rapidly improving and is accelerating. The big tasks are likely a combination of regulations and getting enough data for the machine learning to begin paying dividends in terms of progress. It's very exciting.

 

[Statistical]

I think the 2030 estimate is too far out, machine intelligence capability is rapidly improving and is accelerating. The big tasks are likely a combination of regulations and getting enough data for the machine learning to begin paying dividends in terms of progress. It's very exciting.

I think 2030 for a Level 4 vehicle is realistic. We probably will see retail Level 3 vehicles sooner than that. A Level 3 vehicle can rely on the operator as needed. So if you can make a car which can handle 99% of the situations you can still notify the operator to take over the 1% of the time your system has no idea what to do. A level 4 vehicle is "car take me to work and wake me up when we get there". That is what Uber would need to replace drivers. It has no human to fall back on. It has to handle every situation correctly all the time. The long end of the tail can be a challenging engineering problem to solve.

 

[SLee]

Statistical[/url]":1y0z69q4]

Fee paid by Uber to drivers = driver's capital cost + finance cost + maintenance + fuel + driver compensation.

Sure I agree with that. But a primary difference is that the typical Uber driver uses an older vehicle, anywhere from several to 15 years old. The SDC will be brand new, the key question will be how expensive the self-driving package ends up being.

 

[melgross]

And people argue about where the jobs are going. While some go to China and other countries, a lot of them are gone due to automation.

 

[Janeazy]

my old car got owned by the potholes of Pittsburgh. I wish this guy the best of luck

 

[Corporations Are Patriots]

"Uber, take me to the Chop Shop on S and Twenty. Then close your eyes for about five minutes and you'll get a big surprise!"

 

[MojoSlim]

tetrapyloctomy[/url]":2qdsdw2h]If they don't slow down in the Squirrel Hill Tunnel for no other reason thsn, "Ah! Tunnel!" then they're already better than the average Yinzer commuter.

'ey now ya jag. Yinz never know if da tube is gowen be closed or not. Don' wanna drive inna da side of a mahntin do ya?. And wad if da hole on da other end iddn' open. Sure da potholes need fixed, but I fer one welcome our robotic overlords.

In all seriousness, this has been a long time coming. Uber has already indicated that they will be leasing a large part of the new Almono development in Hazelwood to build a test track for their vehicles. After poaching most of the CMU robotics faculty their intentions were doubly clear. As others noted, Pittsburgh is a mix of horrible roads, potholes, etc, but also gives the challenge of strange one ways, missing ramps, tons of construction to deal with, fake roads (public staircases), bad, but not atrocious, weather, narrow two way streets, and other assorted ridiculousness. It'll be a great location to test the technology out.

 

[JButler]

I just want to see a well thought out reasoning for this 2030 number. Sounds to me like just a wild guess -- they just can't see it happening anytime soon but also don't want to be seen like a clueless naysayer so they pick a number pretty far out but not too far.

 

[berrardo]

Because it's important to figure out how a robocar will respond to a criminal standing in front of it while another criminal points a pistol through the window.

Uber drivers have nothing to worry about for a decade or two. But it's a great business tax-writeoff for Uber ... looks legit on the surface.

 

[b3b0p]

They should have started in Austin. If there is no driver to do background checks on they have no reason to leave.

 

[Gibborim]

DOOManiac[/url]":ws0xrqr1]One step closer to Johnny Cab!

I for one welcome our Johnny Cab overlords.

 

[norton_I]

Of course the driverless cars will be "independent contractors" and Uber won't be liable for any laws they break...

Seriously though, I don't understand how this seems weird for Uber to be doing. Yes, owning your own fleet is a big step up, but it is a step that makes sense. They aren't some tiny start-up any more. They now have plenty of capital and the credibility to raise a lot more if it comes with a solid business plan. And the plan of "undercut the competition on fares while having higher margins by keeping 100%" is about as solid of a plan as you can get.

 

[RoninX]

Statistical[/url]":3qzoh0rs]

I think the 2030 estimate is too far out, machine intelligence capability is rapidly improving and is accelerating. The big tasks are likely a combination of regulations and getting enough data for the machine learning to begin paying dividends in terms of progress. It's very exciting.

I think 2030 for a Level 4 vehicle is realistic. We probably will see retail Level 3 vehicles sooner than that. A Level 3 vehicle can rely on the operator as needed. So if you can make a car which can handle 99% of the situations you can still notify the operator to take over the 1% of the time your system has no idea what to do. A level 4 vehicle is "car take me to work and wake me up when we get there". That is what Uber would need to replace drivers. It has no human to fall back on. It has to handle every situation correctly all the time. The long end of the tail can be a challenging engineering problem to solve.

Yeah, although I could see some issues with the 99% model, if it's not implemented well:

Human: Spends the last 30 minutes reading Facebook updates and chatting with friends.

AI: Sensor failure! Take the wheel, human!

That's probably not going to work.

(Though if the AI can at least pull over safely when it's confused, that might work.)

 

[Wolvenmoon]

This'll work well for them. They won't have any competitors. Especially when I can go out and buy an autonomous vehicle, pull out my phone, type in times that I won't need it, and set it to "Taxi mode" to make a little money on the side while I don't need to use it.

 

[frankdozier]

LordDaMan[/url]":cg0mxe6n]They picked Pittsburgh because the road system is *horrible* here. Streets all willy nilly with no central planning, pot holes, really bad drivers, etc. I know, i drive it every day

If it can drive here and not get into an accident, it can drive anywhere

Not to mention the weather...

I swear I saw one of these fuckers going up Rialto from the 31st street bridge this afternoon. I saw the camera thing on top so I thought it was a mapping car, but I think it had Uber painted on it. I didn't get a good enough look then it was gone.

Anyway, you're completely correct. If that thing can drive in Pittsburgh, then it can drive anywhere. I am completely signed up to drive for Lyft to make extra money, but I seriously don't think I could hack it. If I still lived in Dallas there'd be no problem. But this town is just nuts. NVTS, nuts...

 

[RoninX]

frankdozier[/url]":29ab0iet]

LordDaMan[/url]":29ab0iet]They picked Pittsburgh because the road system is *horrible* here. Streets all willy nilly with no central planning, pot holes, really bad drivers, etc. I know, i drive it every day

If it can drive here and not get into an accident, it can drive anywhere

Not to mention the weather...

I swear I saw one of these fuckers going up Rialto from the 31st street bridge this afternoon. I saw the camera thing on top so I thought it was a mapping car, but I think it had Uber painted on it. I didn't get a good enough look then it was gone.

Anyway, you're completely correct. If that thing can drive in Pittsburgh, then it can drive anywhere. I am completely signed up to drive for Lyft to make extra money, but I seriously don't think I could hack it. If I still lived in Dallas there'd be no problem. But this town is just nuts. NVTS, nuts...

Fortunately both CMU and MIT are located in cities where the drivers are insane. Stanford, on the other hand, may lose out, considering the relatively laid-back drivers in Palo Alto.

 

[Svip]

Gasp, who I am going to be talking to while getting a cab ride then? It's going to be some awkward silence I am going to experience with myself and the car.

Ooh, ooh, will it still have the option of 'follow that car'?

 

[r3loaded]

Svip[/url]":2fb5air5]Gasp, who I am going to be talking to while getting a cab ride then? It's going to be some awkward silence I am going to experience with myself and the car.

Ooh, ooh, will it still have the option of 'follow that car'?

I'm sure that for a small surcharge Uber could add a "movie chase" setting.

 

[my_log]

Statistical[/url]":mz0jqgdt]

I think the 2030 estimate is too far out, machine intelligence capability is rapidly improving and is accelerating. The big tasks are likely a combination of regulations and getting enough data for the machine learning to begin paying dividends in terms of progress. It's very exciting.

I think 2030 for a Level 4 vehicle is realistic. We probably will see retail Level 3 vehicles sooner than that. A Level 3 vehicle can rely on the operator as needed. So if you can make a car which can handle 99% of the situations you can still notify the operator to take over the 1% of the time your system has no idea what to do. A level 4 vehicle is "car take me to work and wake me up when we get there". That is what Uber would need to replace drivers. It has no human to fall back on. It has to handle every situation correctly all the time. The long end of the tail can be a challenging engineering problem to solve.

Level 3 is insane and should be forbidden for obvious reasons.
As for taking bets for 2030, that sounds quite a stretch. Were where we 15 years ago? Smartphones were a fantasy, maps barely existed, voice/image recognition were horrible. Google cars effectively started in 2009 and look where they are now.
It might take some time to be ubiquituous but I can see operational level 4 much sooner.

 

[my_log]

JButler[/url]":2kjijm8r]But they still won't let you overcome the crucial software modeling, (...) The only way out is to drastically simplify the driving condition computers have to deal with. But then it becomes a chicken and egg problem because we can't just lay down a massive Disneyland monorail-type infrastructure everywhere.

This is a deep misunderstanding. Those system seldomly need "software modeling", they rely on machine learning aka "implicit programming" which -supposedly- handles the complexity.

 

[Shavano]

My only surprise is that this is happening so soon. That it would be a cab company pushing out the first real-world application of autonomous vehicles is not a surprise because they have the financial incentive. That it would be Uber is also not a surprise. They're flush with tons of cash that they collected by skirting taxi regulations worldwide. That it's Pittsburgh is a little more of a surprise if only because we don't equate high tech with Pittsburgh. I suspect it's a matter of the regulatory barriers being lower there than other places they might test.

 

[DriveBy]

At least passengers will be less likely to be raped once Uber cars go driverless.

 

[Derecho Imminent]

Does a driverless car require insurance to operate, and have any insurance corps yet committed to providing it?

 

[Statistical]

Does a driverless car require insurance to operate, and have any insurance corps yet committed to providing it?

Nobody knows how that is going to work yet, nobody has even written the regs or even had serious discussions about what the regs will be. Obviously even self driving cars will need some form of insurance. Car insurance covers more than just driver liability.

All "self driving cars" on the road to date have a driver and the insurance and liability is with that driver.