Hacker News Board Hacker News, browsed like a classic message board
Live data from Hacker News

Waymo’s Backseat Drivers: Confidential Data Reveals Self-Driving Taxi Hurdles

theinformation.com

Post reply on HN
Print view View original

101–110 of 196 posts

Re: Waymo’s Backseat Drivers: Confidential Data Reveals Self-Driving Taxi Hurdles

#101

Post by erfgh »

dumbfoundded wrote: post #8
ttul wrote: post #4

The slow progress is not an indictment of self driving. This is one of the toughest engineering challenges ever mounted by humanity.

Yeah, that's a bit of an overstatement. We created nukes, landed on the moon, took sludge out of the ground and used it to power the world. We connected this world with wires and glass fibers to build a real-time global communication system that also gathered all of the world's information in a singular and immediately accessible place. Building a self driving car is hard but really not as tough. How quickly do think…

It's way tougher.

Re: Waymo’s Backseat Drivers: Confidential Data Reveals Self-Driving Taxi Hurdles

#102

Post by zdragnar »

falcor84 wrote: post #69
zdragnar wrote: post #65

Earlier quoted context omitted.

> So far, nobody seems to have full self driving without "safety drivers". A common saying around here is that we have two seasons: winter and (road) construction. Construction zones have pretty much every obstacle to automated driving you can think of: * painted lanes that don't correlate to the temporary lanes marked by cones * lanes that don't correspond to pre-programmed maps / gps * irregular and unpredictable v…

>Oh, and said protocol has to be hack-proof so trouble-makers can't start convincing cars that they're in the middle of a construction zone and force them out of their lanes on normal roads. Why would it be easier for trouble-makers to fool autonomous cars? As a human driver, I'd be fooled by pretty much any road marking or guy in an orange vest.

> pretty much any road marking or guy in an orange vest

That only works if a police car doesn't come by and catch the perpetrator in the act.

With a wireless communication to automated drivers, someone could plausibly feed bad information from a hidden or otherwise remote location.

Beyond that, just as automation allows human-intensive processes to scale by removing the humans, fooling automated drivers can scale much more readily than fooling human drivers.

Re: Waymo’s Backseat Drivers: Confidential Data Reveals Self-Driving Taxi Hurdles

#103

Post by Animats »

daniel_iversen wrote: post #89
Animats wrote: post #54

That's useful info. I wonder where they had trouble with erratic steering. One Cruise video shows that where driving is alongside parked cars on a narrow street and the parking is irregular. The self driving vehicle is trying to stay in its half of the street unless absolutely necessary. Waymo still has to use "safety drivers". So far, nobody seems to have full self driving without "safety drivers". Which makes it us…

Mary Meeker's internet report this year is one place where you can see a summary chart that suggests that not only is technology innovation happening exponentially (which a lot of people agree to) our adaptability is also growing (which I guess it has to in order for innovation growth to continue like that?): https://www.bondcap.com/report/itr19/#view/156 (however what's most interesting about that slide is that we n…

That's not data. No scale, no data.

Actually, the period of greatest technical change in human history was probably 1860-1910. Steel, electricity, autos, airplanes, subways, radio, elevators, skyscrapers, machine guns...

Re: Waymo’s Backseat Drivers: Confidential Data Reveals Self-Driving Taxi Hurdles

#104

Post by Nokinside »

zdragnar wrote: post #65
Animats wrote: post #54

That's useful info. I wonder where they had trouble with erratic steering. One Cruise video shows that where driving is alongside parked cars on a narrow street and the parking is irregular. The self driving vehicle is trying to stay in its half of the street unless absolutely necessary. Waymo still has to use "safety drivers". So far, nobody seems to have full self driving without "safety drivers". Which makes it us…

> So far, nobody seems to have full self driving without "safety drivers". A common saying around here is that we have two seasons: winter and (road) construction. Construction zones have pretty much every obstacle to automated driving you can think of: * painted lanes that don't correlate to the temporary lanes marked by cones * lanes that don't correspond to pre-programmed maps / gps * irregular and unpredictable v…

Remote operators solve these issue.

Once you have autonomous cars that driving safely, but can't manage complex situations like you describe, you delegate those for remote pilots that are allowed to operate car in slow speeds. You need 5G network coverage with mission-critical features (mcMTC) to archieve that. BLER 10^-6 and E2E latency Taxi fleet of 10,000 vehicles might need only 100-200 remote operators to manage the fleet. That kind of reduction in workforce provides huge savings.

Re: Waymo’s Backseat Drivers: Confidential Data Reveals Self-Driving Taxi Hurdles

#105

Post by alkonaut »

dumbfoundded wrote: post #8
ttul wrote: post #4

The slow progress is not an indictment of self driving. This is one of the toughest engineering challenges ever mounted by humanity.

Yeah, that's a bit of an overstatement. We created nukes, landed on the moon, took sludge out of the ground and used it to power the world. We connected this world with wires and glass fibers to build a real-time global communication system that also gathered all of the world's information in a singular and immediately accessible place. Building a self driving car is hard but really not as tough. How quickly do think…

My guess is that it (full 100% self driving) has around the same technological difficulty as putting a man on Mars. It also depends on what constitutes "100%" of course.

> How quickly do think we'd get self driving cars if the USA spent 4% of the federal budget on it like NASA received in the 60s. (That's about $40B a year for a decade).

At that cost we could adapt all infrastructure to suit self driving cars, instead of developing self driving cars to adapt to human infrastructure. But I think that kind of cost is always going to be beyond what's acceptable.

I think the discussion is mostly pointless because of diminishing returns: if you can have "99.9% full self driving" for a tiny fraction of the cost, who would want to pay to go from 99.9% to 100%?

Initially human remote drivers will take care of the rest. And then there is a very slow commercial race towards using fewer humans that drives the very slow march to 99.9% and 99.99% self driving and so on. Driving the last second of the last edge case route is basically something that requires AGI (as long as we don't adapt infrastructure).

Re: Waymo’s Backseat Drivers: Confidential Data Reveals Self-Driving Taxi Hurdles

#106

Post by IshKebab »

dmix wrote: post #95

> The data provide the most detailed view yet of the passenger experience inside vehicles from Waymo, which doesn’t disclose the information even though it uses public roads as its testing grounds. Why would it riding on public roads matter when it's about what happens inside a private vehicle during a high-value R&D experiment? They got permission from Arizona to run the tests and it will likely benefit Arizona's ec…

Yes! The public definitely has an interest in how dangerous other cars on public roads are. Imagine if the data showed that they'd had tons of near misses with cyclists for example.

Re: Waymo’s Backseat Drivers: Confidential Data Reveals Self-Driving Taxi Hurdles

#107

Post by alkonaut »

moomin wrote: post #26

This might just be me, but it strikes me that the biggest problem self-driving cars has is that no-one actually needs them. Don’t get me wrong, you try to sell me a car that has the (fully capable) feature I’d definitely want it. But need it? Well, you’re not going to solve the problem of too many cars and the slow traffic that results with cars. It’d be hard to get special treatment for cars with a premium feature.…

So long as I need to watch it, I don't feel a huge need for my personal car to be self driving. It doesn't free up any time for me, so I'm not ready to pay for it.

But if we had "full" self driving, as in the car can drive empty, that opens up new possibilities.

First of all, I could avoid owning a car all together. I could be part of a car sharing/pool system where I can summon a car when I want it instead. Those pools exist but their biggest drawback is you have to get TO the car instead of having it on your driveway.

Second, I could get driven home after having a drink. Where I live the legal limit is 0.02 so one drink equals a Taxi.

Those are features I'd very much be willing to pay for.

Re: Waymo’s Backseat Drivers: Confidential Data Reveals Self-Driving Taxi Hurdles

#108

Post by IshKebab »

mdorazio wrote: post #2

Interesting analysis using comments/rider scores. 10% improvement over 6 months is actually pretty good, though. Even if that rate slows as trickier edge cases get handled, It still implies early 2020s for true driverless readiness in select locations. That's pretty much on track with overall predictions for the autonomous market.

That's assuming their rate of improvement is constant and there isn't a long tail of difficulty. But there is. The long tail of difficult driving situations is very very long.

An example from my most recent drive: I drove through a common where cows graze. I doubt a driverless car is programmed to slow down near cows.

The only solution I can see is to whitelist roads and start with the simplest (motorways/highways) then gradually expand to more and more roads. I guess that's kind of what they're doing - suburban America is very easy (although you still have pedestrians and cyclists to deal with unlike motorways), and maybe whitelisting is the cause of the routing complaint.

Re: Waymo’s Backseat Drivers: Confidential Data Reveals Self-Driving Taxi Hurdles

#109

Post by DiogenesKynikos »

falcor84 wrote: post #70
scarmig wrote: post #45

Earlier quoted context omitted.

5 stars isn't perfect. 5 stars is "there's nothing so bad that I want to take the time to complain about it." Waymo employees, who are encouraged to be especially tough, give reviews that are 47% negative. That's likely closer to the metric for perfect. We don't get perfection from human drivers either, of course. Though part of the promise of Waymo is much better performance than humans, which it apparently isn't cl…

If I were "encouraged to be especially tough", I can promise you that the taxis I ride in would get a much worse score than 47% negative.

For entirely different reasons, though. Your taxi driver is not going to randomly freak out because they mistook a bush for a pedestrian or didn't understand what traffic cones mean. The bad taxi driver is just going to drive too fast while talking on their cellphone.

Re: Waymo’s Backseat Drivers: Confidential Data Reveals Self-Driving Taxi Hurdles

#110

Post by leoc »

Nokinside wrote: post #104
zdragnar wrote: post #65

Earlier quoted context omitted.

> So far, nobody seems to have full self driving without "safety drivers". A common saying around here is that we have two seasons: winter and (road) construction. Construction zones have pretty much every obstacle to automated driving you can think of: * painted lanes that don't correlate to the temporary lanes marked by cones * lanes that don't correspond to pre-programmed maps / gps * irregular and unpredictable v…

Remote operators solve these issue. Once you have autonomous cars that driving safely, but can't manage complex situations like you describe, you delegate those for remote pilots that are allowed to operate car in slow speeds. You need 5G network coverage with mission-critical features (mcMTC) to archieve that. BLER 10^-6 and E2E latency Taxi fleet of 10,000 vehicles might need only 100-200 remote operators to manage…

> You need 5G network coverage with mission-critical features (mcMTC) to archieve that. BLER 10^-6 and E2E latency I do wonder if that's a factor behind Musk's push into low-orbit satellite Internet.

> Taxi fleet of 10,000 vehicles might need only 100-200 remote operators to manage the fleet. That kind of reduction in workforce provides huge savings.

Even if all mileage was human-driven there would be very large benefits if you could really consolidate taxi drivers in call-centres for remote driving. No need to transport or preposition drivers and much less trouble estimating demand.

Post reply on HN