Waymo’s driverless cars have logged 10M miles on public roads | Hacker News Reader

Waymo’s driverless cars have logged 10M miles on public roads

qz.com

40 pointsNicoJuicy40 comments

> Waymo cars reportedly have problems turning left, a flaw the Information calls “the Zoolander problem”: local residents reported waiting behind Waymo cars for minutes at a time as the algorithm hesitates to turn into traffic.

UPS drivers are told not to turn left for a reason. Couldn't Waymo do the same?

I think that only works for UPS because everyone else is turning left. But imagine a world where no car will turn left. I wonder if traffic would flow better or not.

I suspect not, because if that were the case, cities would outlaw left turns, and as far as I know, no city has done that.

Fair enough.

There is a difference between individual behavior and the law.

Imagine a world where no person drinks alcohol. I wonder if it would be more peaceful.

I suspect yes. Even though outlawing alcohol is not a solution that would lead to this world.

The issue is that "avoid left turns if possible" adds extra mileage and extra traffic on those streets that are used for a set of right turns; plus it's a SHOULD, not a MUST [RFC2119] - sometimes the right-turn detour would be just too long. UPS has a slight advantage here as it has a set of points to visit, not necessarily in a given order - more routing options, as compared to an A-to-B route.

> But imagine a world where no car will turn left. I wonder if traffic would flow better or not.

We have a ton of roundabouts in Europe and if you ignore the merge into the roundabout, it transforms any turn into a right turn.

I'm a huge fan of roundabouts, and wish we had more (or any) of them here in the states. But barring redoing the pavement at every intersection, that's not a very viable plan here. :)

They come in various sizes. In some extreme cases the circle is completely inside the crossing of the roads and doesn't even require any change to the existing roads. Like here: https://www.google.com/maps/@44.8481471,-0.6336833,3a,75y,18...

If you can't turn left, it's the end of NASCAR! ;)

It should. Sometimes I choose to do the same thing if the line of cars gets too long behind me and I don't want to waste their time. Waiting when a human would have turned left should count as a disengagement.

I've seen a huge uptick of Waymo cars on the road lately. I live in Cupertino (the next town over from Mt. View) and yesterday I saw three within the span of one light cycle, each going in a different direction. At this point I see at least one if not a lot more every weekday.

In your experience do they drive differently than any other car on the road? If a vehicle around you was powered by Waymo with no markings stating as much would you know (ignoring the fact of an empty driver's seat)?

> In your experience do they drive differently than any other car on the road?

For sure. They drive like the most conservative driver you'll meet, to the point of being a bit dangerous. They come to a full stop at every stop sign, which here in California is unexpected, and unexpected means dangerous. They also slam on their brakes a lot, without regard to how far away the person behind them is (as opposed to when I have to slam on the brakes, I check the rear view and if possible cut it a little closer to the car in front if the car behind is really close).

They putter out of stop lights, which is mostly just annoying, and they generally follow the speed limits, even on roads where all the rest of the traffic is typically 10mph over.

To be fair, they are doing what in theory is much safer, but since they don't act like a human, it makes them unpredictable.

> If a vehicle around you was powered by Waymo with no markings stating as much would you know (ignoring the fact of an empty driver's seat)?

Well the massive LIDAR unit on top is a big giveaway. :) Actually that's the biggest giveaway. The cars are unmarked otherwise, and there is always a driver in the seat. They just may not have their hands on the wheel, but that's hard to tell from outside. And of course all of the aforementioned strange driving habits.

> "Competing with other companies with autonomous-vehicle programs like Uber, Tesla, Apple, and GM’s Cruise, Waymo is leading the pack in terms of road miles driven."

I'd think Tesla would be leading in terms of autonomous mikes driven given that they have sold 100k+ vehicles with Autopilot.

Does Tesla Autopilot navigate controlled intersections on it's own?

If not, it's kind of a silly comparison.

It doesn't. It's just RGB camera + front-facing radar based lane keeping and traffic aware cruise control. It will also change lanes when you signal, but that's just augmented lane keeping.

It's frustrating and dangerous how many people don't realize the massive difference there is between Autopilot and autonomy. It's mostly Tesla's fault in how they market it IMO.

No, if you want to compare autonomous miles driven by that definition the Chevrolet Cruze is leading in terms of autonomous miles driven given that they sold the most vehicles in the US with cruise control.

If you want to talk about real autonomous driving then it's Waymo.

Do Cruze vehicles supply data back to GM constantly that is used to improve self driving models? Updated models which are pushed back down to vehicles?

Can these systems reason? If a driverless car went on a road that started spiraling so that the road would be upside down, would the car just keep following the road, not realizing it will fall down or would it realize it cannot continue?

I'd bet humans would handle a corkscrew road that makes the car do a barrel roll (I think that's what you're trying to describe) worse than a robot would. Luckily that type of road doesn't exist, for numerous reasons that have nothing to do with robots.

I know it's a serious conversation but...

> Luckily that type of road doesn't exist

Speak for yourself.

In more related news I have a friend who helps out on Waymo and he's ALWAYS posting great articles about Waymo tech, always showing off amazing pictures of what he's working on, and he's always positing about how great things in the future could be. He's not the type of dude to over hype something, he's not the type of person who over exaggerates things either. He's not insanely fanboy-ish about it but he's definitely borderline. But, I guess when your vehicles are blowing the competition out of the water you'd be excited too?

There will be fun attacks/pranks on driverless cars. Imagine waiting for it to enter a roundabout but your friends are at all the exits holding big wrong way / no entry signs.

But then the car has full video footage of everything your doing to hand over to the police...

When did the meaning of the adjective "driverless" change from "without a driver" to "with a safety driver, engineer and a remote operator"?

Where did they say that anywhere here?

It is quite clear from recent press reports that the vast majority of Waymo's 10M miles were logged with a safety driver, and some with an additional safety engineer, but the title talks about "Waymo's driverless cars".

So obviously, the adjective driverless no longer means without a driver.

The part about an additional remote operator is tongue in cheek.

https://www.theinformation.com/articles/waymos-big-ambitions...

It doesn't say 10M miles by how many cars though? Could they start trialing discounted Taxi Ride on a massive scale? How many more miles of testing do we need before this can be rolled out commercially?

The USA has 7.1 fatalities per billion km driven.

Thats 1 fatality per ~100M miles on average.

So presumably Waymo would have to drive a few hundred million miles without a fatality to conclusively be able to say 'we are safer'.

You're assuming that the only thing an accident can result in is loss of human life. Injury statistics are far more interesting because they occur more frequently. [1] At a rate of 24 injuries per 100 million km waymo cars should have resulted in quite a few injuries already.

[1] https://www.transport.govt.nz/resources/tmif/safetyandsecuri...

Waymo has already had some injuries.

> There was an occupant in the driver's seat of the Waymo vehicle who suffered minor injuries, Tyler added.

https://phys.org/news/2018-05-waymo-self-driving-car-collisi...

A human drove into a Waymo car. I would not classify that as Waymo having some injuries. The driving safety statistics should only cover decisions made by the auto-pilot. If not simply because the same auto-pilot will be used in the future but not the same vehicles / vehicle seating arrangements and related safety measures.

Yeah but what about just crashes per mile? I'm sure there's more than plenty enough data to determine if Waymo is better in at least that. Since you need a crash to have a fatality, safety comparisons can be inferred from less than 3 million miles.

It should be possible to work that number quite a bit lower.

Insurance companies are (literally) in the business of pricing the riskiness of drivers, so they tend to understand which ones are the riskiest. If there is a group that is clearly riskier than the total average but is allowed to drive, there's good reason to think about making that group the point of comparison.

I mean they could have 100x more cars testing it and it would be 100M miles in little time. The problem is that is going to cost quite a bit. In a perfect world Google would sink in additional few billions and massive scale it for testing. So we could technically have Waymo taxi in 3 - 4 years time.

Here is the pace.

2012 miles per day 240

2014 miles per day 636

2017 miles per day 4,630

Feb 2018 miles per day 10,753

June 2018 miles per day 16,494

July 2018 miles per day 25,000

Interestingly enough, the new version is supposed to have a completely revamped sensor suite including a new LIDAR, new cameras and radar.

This suggests two interesting things to me:

1) Google/Waymo, the "software and ML can solve this problem" company is admitting that hardware is important to solving the problem.

2) They are likely much further away from L5 or even public release than some believe given that they are now completely revamping their sensor suite.

Not sure where you got the idea that Waymo doesn’t value hardware development - their whole lawsuit saga against Uber was over lidar trade secrets (google “laser is the sauce”). Maybe you were thinking of Tesla who has insisted their existing video/radar stack will be competitive with enough ML? For all the other major SDV companies I’m familiar with, affordable and peformative lidar is a huge focus.

Waymo handed out pcbs as earrings and their patents were bs.

Most companies want lidar to get cheaper not better. Very rarely has the solution to a perception problem been "better sensors" as opposed to "more ml".

I interpreted it as the exact opposite :)

If Waymo can revamp the car's sensor suite this late in the process, all the while ramping up their fleets and consumer services, it suggests to me that their software is sufficiently powerful and flexible not to need massive and costly retraining; basically I think they have "solved" the problem on the software side (given appropriate constraints such as good weather, excellent mapping, etc.) and are now working on rationalizing the hardware costs.

I think it's safe to say they wouldn't be logging anywhere near 10M miles if that weren't the case.

Especially when the graph shows that they've completed 6M of those 10M in the last year! (https://www.theatlas.com/charts/S1uAcpj97)

Reply on news.ycombinator.com