Waymo's Driverless Cars Have Logged 10 Million Miles On Public Roads (qz.com)

An anonymous reader quotes a report from Quartz: Alphabet's driverless-car company Waymo announced a new milestone today (Oct. 10): its vehicles have driven a collective 10 million miles on U.S. roads. With cars in six states, Waymo has really been racking up the miles since April 2017, when it launched a program giving rides to passengers around the Phoenix, Arizona area. At that point, Waymo cars had driven not quite 3 million miles since the company's earliest days as a research project within Google in 2009. But in the last 18 months, the company more than tripled its road mileage.

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. [...] The company's next 10 million miles, CEO John Krafcik said in today's announcement, will focus on "striking the balance" between its safety-first algorithms and driving assertively in everyday maneuvers, like merging, and navigating bad weather. But it's worth keeping things in perspective: U.S. drivers rack up some 3 trillion miles each year, so Waymo still has some ground to cover.

Comments

covering ground being the operative word

[Austerity Empowers]

Even if they put 3 trillion miles on their system, if they confine it to just a few geographical areas, I don't trust it very much. I'd like to see them driving in NYC, Boston, Chicago, New Jersey (even humans can't figure this one out), etc. Places where public investment in the roadways has either been compromised (i.e. stolen by politician for other bullshit), minimal, or there simply wasn't enough space to put proper roads in, so they did something else instead...

Re:covering ground being the operative word

[religionofpeas]

Don't worry. Every slashdot comment is framed and hung up in the board room. True goldmine here.

Re:covering ground being the operative word

[phantomfive]

Waymo's system can't operate in an area where they haven't built a highly detailed 3D map. NYC isn't dramatically worse than San Francisco (which has plenty of bizarre traffic things, but it doesn't matter because the AI has a really good map. It knows what those things are), and Waymo has been operating in SF. If they can build the map, they can handle NY or Boston ok.

Re:covering ground being the operative word

[religionofpeas]

Small changes in the environment shouldn't matter. In the future, they could automatically make updates to the map using the 3D scans from all the cars passing points that show discrepancies in the old map. Maybe they're already doing that.

Re: covering ground being the operative word

[peragrin]

Then you need to Learn the differences in self driving systems.

Waymo is the only level 4 company and has tight geofenced areas, and high res maps.

Level 0 is regular cruise control
Level 1 is the newer speed changing in traffic cruise control
Tesla and supercrusie are level 2

Fully self driving starts at level 4 but limited
Level 5 car drives like humans can. No one has even started this yet.

It is why I laugh when people say self driving cars are almost here. Nope 20-30 years away at best.

Re:covering ground being the operative word

[morethanapapercert]

You make a good point about scientific experiments normally requiring Informed Consent from its participants. But I have three counter-balancing points for you to consider:

1) There are a number of people out there who I think should not be driving. I don't consent to them sharing the road with me. But I am forced to accept that I must share the road with them because it is public infrastructure. Any member of the public who qualifies based on a pretty easy driving test is allowed to use the roads in a motor vehicle. Based on that idea, an autonomous vehicle that is demonstrably as safe or safer than your average driver in a road test should be allowed to access the public roads.

2) People with bikes, scooters, roller blades et al don't even need a license. It is akin to the concept of the commons, areas where anyone could bring their herds to pasture and/or people could cross freely on their way to somewhere else. Choosing to use that area when headed to my destination is tacit acceptance of the risk that I will step in manure or be confronted with animals who interfere with my progress. Likewise, choosing to use the public roads can be construed as acceptance that there is a certain irreducible level of risk in doing so. The question isn't whether autonomous vehicles represent a risk to the general public. It is whether it measurably increases the level of risk that is already there.

3) No matter how much closed track testing you do, no matter how complex and varied your test environment is, sooner or later it will be deployed on public roads. No matter how much preparation you have done. that initial deployment is still technically an experiment. As with drug testing, no matter how careful we are in the testing phase, there is still a chance that we will uncover one or more flaws, potentially serious flaws once it is available for general prescribing. What is important is having an effective process in place where those flaws can be quickly identified, confirmed and mitigated or solved.

AI

[Artem S. Tashkinov]

Every time when I hear about these crazy miles I've got just one question to ask: has Google finally solved image recognition and I'm not talking about simple cases - I'm talking about deliberate fakes, bad weather conditions, etc. 1, 2, 3.

These issues can easily make your car software make life threatening decisions.

Re:Huh?

[XXongo]

Um, huh? Tesla's Autopilot had driven 1,2 billion miles as of July. Two orders of magnitude more than Waymo.

Uh, Tesla's "autopilot" is a driver assist, not a self-driving vehicle. And it racks up the miles on expressways-- that's the easy kind of driving.

So, no, not the same thing.

10 million miles is really nothing. In the US, there's only one fatal accident per 86 million miles on average.

Indeed, that's the metric to compare to. But not all miles driven are the same.

Perspective?

[sjbe]

But it's worth keeping things in perspective: U.S. drivers rack up some 3 trillion miles each year, so Waymo still has some ground to cover.

Umm, WTF does this have to do with "keeping perspective"? It isn't a competition between Waymo and the rest of us human drivers to see who can rack up the most miles driven.

Re:Perspective?

[ledow]

It means that non-human-driven cars have collectively accounted for 0.000333% of all the miles travelled this year.

This means that... pretty much... they haven't way over 99.99% of the human race has never been in, with or near a driverless Waymo car.

This means that, pretty much, Waymo hasn't even covered a thousandth of a percent of the things it needs to cope with when driving.

And also... that if there is a single accident, that it would scale up 300,000-fold in terms of their overall average accident rate if we all jumped on board them. And Waymo's had quite a few.

Much more interesting a statistics would be: How much of this was at speed, on highways, etc. Because highways are much easier to drive on for automated cars (and humans, if they are sensible). But for sure, if you can make an autonomous car than can navigate the Hangar Lane Gyratory in London it would mean a whole lot more than even a million miles of highway driving.

Humans are (often) shitty drivers

[sjbe]

Even if they put 3 trillion miles on their system, if they confine it to just a few geographical areas, I don't trust it very much.

Why? Human drivers are demonstrably dangerous and the body count to date heavily favors the computers as the likely safer option. While I'm not suggesting autonomous driving vehicles are ready for prime time yet or that it's a slam dunk that they are safer, I think people like yourself are not really doing a very good job of evaluating the actual risk data. Honestly I don't really trust YOU as a driver either. Nothing personal - you shouldn't trust me either or any other human driver. But the point is that what you should trust is the data and the data so far seems promising with regards to the safety of these things.

Re:Trust

[fluffernutter]

Waymo drives 0.0003 percent the miles humans drive a year in hand-picked conditions and you talk like they deserve a safety award or something.