Slashdot Mirror


Waymo's Autonomous Vehicles Are Driving 25,000 Miles Every Day (techcrunch.com)

With Nevada Governor Brian Sandoval at the National Governors Association, Waymo CEO John Krafcik announced a huge milestone: Waymo's fleet of self-driving vehicles are now logging 25,000 miles every day on public roads. The company reportedly has 600 self-driving Chrysler Pacifica Hybrid minivans on the road in 25 cities. Waymo has also driven 8 million miles on public roads using its autonomous vehicles, "meaning the comopany has been able to double the number of autonomous miles driven on public roads in just eight months," reports TechCrunch. From the report: The company also relies on simulation as it works to build an AI-based self-driving system that performs better than a human. In the past nine years, Waymo has "driven" more than 5 billion miles in its simulation, according to the company. That's the equivalent to 25,000 virtual cars driving all day, everyday, the company says. This newly shared goal signals Waymo is getting closer to launching a commercial driverless transportation service later this year. More than 400 residents in Phoenix have been trialing Waymo's technology by using an app to hail self-driving Chrysler Pacifica Hybrid minivans. The company says it plans to launch its service later this year.

58 of 101 comments (clear)

  1. That's fast by slinches · · Score: 1

    I didn't think those minivans could go over 1000 mph.

    --
    Knowledge Brings Fear
    1. Re:That's fast by Tablizer · · Score: 1

      I didn't think those minivans could go over 1000 mph.

      They can when falling off cliffs.

    2. Re:That's fast by Tuidjy · · Score: 1

      Nope. They are not aerodynamic enough for that. Falling off a cargo plane, maybe. Falling off a balloon in the stratosphere, definitely.

      --
      No good deed goes unpunished...
    3. Re:That's fast by omnichad · · Score: 1

      It would weigh roughly 1200 tons at that size. Even if you coiled it up, you could never travel far.

    4. Re:That's fast by thsths · · Score: 1

      Exactly why thought. Why are Slashdot headlines always incorrect?

      Slashdot - fake news for fake nerds

  2. My 100% correct prediction about these machines by hyades1 · · Score: 1

    The most frequent users will be hookers. No question.

    --
    I've calculated my velocity with such exquisite precision that I have no idea where I am.
    1. Re:My 100% correct prediction about these machines by CaptainDork · · Score: 1

      I've calculated my velocity with such exquisite precision that I have no idea where I am.

      Your sig wins.

      --
      It little behooves the best of us to comment on the rest of us.
    2. Re:My 100% correct prediction about these machines by hyades1 · · Score: 1

      Thank you, good sir.

      --
      I've calculated my velocity with such exquisite precision that I have no idea where I am.
  3. Really impressive by 110010001000 · · Score: 1

    I was wrong. AI and automation are right around the corner! I really can't wait to get my autonomous car, or even better, to hail one!

    1. Re:Really impressive by gweihir · · Score: 2

      Sorry to disappoint, these things are not "AI". They are just a result from around 50 years of dedicated, specialized automation research. And automation is a very, very old thing.

      --
      Most ACs are not even worth the keystrokes to insult them. Be generically insulted by this and ignored otherwise.
    2. Re:Really impressive by AmiMoJo · · Score: 1

      They use neural nets for image recognition. It's got some AI in it.

      --
      const int one = 65536; (Silvermoon, Texture.cs)
      SJW, n: "Someone I don't like, and by the way I'm a fuckwit" - AC
    3. Re:Really impressive by gweihir · · Score: 1

      Statistical classifiers are not AI, whether trained or parametrized.

      --
      Most ACs are not even worth the keystrokes to insult them. Be generically insulted by this and ignored otherwise.
    4. Re:Really impressive by RespekMyAthorati · · Score: 1

      According to you. Why would anybody care how you define "AI"?
      The "A" stands for "Artificial", not "Authentic".

  4. 'Simulations' mean NOTHING by Rick+Schumann · · Score: 2

    Neither does toodling around at 25mph or less.

    1. Re:'Simulations' mean NOTHING by drinkypoo · · Score: 1

      Do you have any idea what the average speed in town is? I'll bet it's less than 25. Also, miles driven in simulation do not help test the sensors, but they do help test the driving logic.

      --
      "You're right," Fisheye says. "I should have set it on 'whip' or 'chop.'"
    2. Re:'Simulations' mean NOTHING by K.+S.+Kyosuke · · Score: 1

      If you don't care, why the bitching about simulations and speeds?

      --
      Ezekiel 23:20
    3. Re:'Simulations' mean NOTHING by drinkypoo · · Score: 1

      At their current driving rate, it would take them over 130 years to match Tesla's active AP miles and double that to match its shadow mode miles.

      Autopilot doesn't do what their system does. Tesla could have infinitely more miles and still be behind, because their training is for a system with lesser capabilities.

      --
      "You're right," Fisheye says. "I should have set it on 'whip' or 'chop.'"
    4. Re:'Simulations' mean NOTHING by gweihir · · Score: 1

      Actually, almost all tricky situations are at low speed.

      --
      Most ACs are not even worth the keystrokes to insult them. Be generically insulted by this and ignored otherwise.
    5. Re:'Simulations' mean NOTHING by mlyle · · Score: 1

      > is only 4 times more than Tesla's real-world miles on active mode and 2 times more than Tesla in shadow mode.

      > take them over 130 years to match Tesla's active AP miles and double that to match its shadow mode miles.

      Is one of these reversed?

    6. Re:'Simulations' mean NOTHING by thegarbz · · Score: 1

      Neither does toodling around at 25mph or less.

      You mean the scenario that the vast majority of people find themselves in, the vast majority of traffic hazards occur in, and the vast majority of road variance exists in?

      Yeah I'm sure they're much better driving straight lines down highways! /sarcasm.

    7. Re:'Simulations' mean NOTHING by Rick+Schumann · · Score: 1

      Because I'm intellectually offended at SDC fanbois who swallow the media hype and other associated bullshit when I know damned well the entire approach to AI they're using for this and many other things is no different than it was in the 90's and is clearly not up to the task, should not be allowed to be used where human lives will be lost because of it. You can believe all the hype and bullshit all you like but I'm not fooled.

  5. amazing what tesla has taught the industry. by WindBourne · · Score: 2

    Hopefully, Waymo will catch up to Tesla and start giving them some competition.

    --
    I prefer the "u" in honour as it seems to be missing these days.
    1. Re:amazing what tesla has taught the industry. by AbRASiON · · Score: 1, Insightful

      I seriously can't tell if this post is sarcastic or not?

    2. Re:amazing what tesla has taught the industry. by WindBourne · · Score: 1

      What? You think that getting to 100% self-drving cars is not going to involve any accidents?
      The fact is, that Tesla has far fewer accidents / mile driven, than the other APs. BUT, to get to this level, the other APs will have to allow lots of AP based driving. Waymo is finally adding some mileage on their network which is what it will take to teach theirs.

      --
      I prefer the "u" in honour as it seems to be missing these days.
    3. Re:amazing what tesla has taught the industry. by WindBourne · · Score: 1

      both.
      That is the only way that these AP systems are going to learn about edge cases. Right now, MB, Volvo, GM, etc have been pretenders because they have no real mileage on their systems. Until a sytem is putting on 100K miles EVERY day AND jump all over the accidents quickly, they are not going to really develop what is needed.

      The weird thing is, that so many ppl believe that getting ahead can be done without loss of lives, on things like Space or automated car driving. Nothing could be further from the truth. Yet, that does not mean that these are NOT safer than other forms of space access or manual car driving.

      --
      I prefer the "u" in honour as it seems to be missing these days.
    4. Re:amazing what tesla has taught the industry. by citizenr · · Score: 1

      Crashing into huge, visible and stationary obstacle is NOT an edge case, its a design defect.

      --
      Who logs in to gdm? Not I, said the duck.
    5. Re:amazing what tesla has taught the industry. by AbRASiON · · Score: 1

      You are seventh level delusional.

      What a waymo vehicle can do autonomously vs a Tesla doesn't even compare. It's just not even a competition.

    6. Re: amazing what tesla has taught the industry. by WindBourne · · Score: 1

      You are the delusional one. Tesla has over 1000x as much mileage as waymo. And waymo keeps most of their driving on the easy and safer stuff. https://www.google.com/amp/s/w...

      --
      I prefer the "u" in honour as it seems to be missing these days.
    7. Re: amazing what tesla has taught the industry. by crimson+tsunami · · Score: 1

      Just make these a bit bigger and we can put saddles on them and have robot horses. It's the future I tell ya.

  6. Fully autonomous or sort of autonomous? by Nkwe · · Score: 2

    Are these miles fully autonomous (meaning no human in the car) or are the miles with assistance (human in the car, ready to take over or "assist")? If with assistance, what is the rate of help the cars get on some sort of statistic that can be compared over time? (Maybe "assists" per 1000 miles or something like that.) The statistic in the headline sounds impressive, but is it?

    1. Re:Fully autonomous or sort of autonomous? by complete+loony · · Score: 3, Insightful

      Waymo autonomous cars drove 352,545 miles in California between December 2016 and November 2017, and disengaged from autonomous mode 63 times

      No idea if the car was about to do something dangerous, or was disengaged because the driver got impatient. If the car chose to disengage, that's a small enough number that in areas with good data coverage, a driver in a single call center could take over remotely.

      --
      09F91102 no, 455FE104 nope, F190A1E8 uh-uh, 7A5F8A09 that's not it, C87294CE no. Ah! 452F6E403CDF10714E41DFAA257D313F.
    2. Re:Fully autonomous or sort of autonomous? by David_Hart · · Score: 2

      Are these miles fully autonomous (meaning no human in the car) or are the miles with assistance (human in the car, ready to take over or "assist")? If with assistance, what is the rate of help the cars get on some sort of statistic that can be compared over time? (Maybe "assists" per 1000 miles or something like that.)
      The statistic in the headline sounds impressive, but is it?

      It's also missing details such as weather, temperature, types of roads (i.e. pavement vs gravel), how well the roads are marked, traffic conditions, time of day, construction/road changes, amount of pedestrians/cyclists, etc.

    3. Re:Fully autonomous or sort of autonomous? by gweihir · · Score: 2

      This is not the "proof of fitness" stage, this is still the "research" stage, so with human and, if done right, with additional safety features. This does mean that these systems can drop back to human control even when the automatics are just a bit concerned, but could actually handle the situation.

      --
      Most ACs are not even worth the keystrokes to insult them. Be generically insulted by this and ignored otherwise.
    4. Re:Fully autonomous or sort of autonomous? by Gavagai80 · · Score: 1

      Seems to me that the average human driver does something incredibly stupid indicating a serious brain freeze more often than every 5600 miles.

      --
      This space intentionally left blank
    5. Re:Fully autonomous or sort of autonomous? by Anonymous Coward · · Score: 1

      No, it's not... They still disengage the autopilot about every 5000 miles. That's abysmal compared to a human driver. Would you drive with a person that stops their car every 5000 miles with 'sorry can't go on here'? And this is on well marked and mapped roads. How about roads with no markings, single lane, or gravel roads that are on no map? Or roads with worn markings? Or roads where the map and reality don't match? Contruction zones come to mind.

      The 80/20 rule still applies. The first 80% take 20% of the time, the last 20% take 80% of the time.

      Considering what a true self driving car will need to do, we're close to reach the 80% mark.

    6. Re:Fully autonomous or sort of autonomous? by Anonymous Coward · · Score: 1

      No, it's not. Why? Because the hypothetical driver in that call center would have to familiarize himself with the situation first. That can take a bit, in the meantime the car would sit in traffic and block the road in the best case... or hit something in the worst case.

      This 'we let the autopilot drive and a human takes over when it can't deal with things' works in situations where you can freeze things (like a simulation on a computer), but it doesn't work in a moving vehicle where things can go from OK to dangerous to someone is dead in a split second.

    7. Re:Fully autonomous or sort of autonomous? by thelUrk · · Score: 1

      Would you drive with a person that stops their car every 5000 miles with 'sorry can't go on here'?

      My car only goes about 300 miles before it will stop all by itself anyway. If the driver isn't smart enough to find a safe place to pull over, get out, walk around a bit, maybe refill the fuel tank, then pretending the number "5000 miles" is significant is just silly. All that means is that is how often it takes for them (on average I'm guessing) to get stumped by something they haven't learned yet.

      My youngest just got a learner's driving permit. In my experience from teaching young new human drivers, it's really not that different. In the beginning they may make several mistakes per mile. Within a few days or maybe even hours they catch on and the rate drops to maybe one error per session of 5-30 miles. Some states require having a permit and accompanying driver for a full year, though I would feel comfortable letting them go on their own in as little as 3 months. (Disclaimer: I learned to drive as a CDL professional and teach that style to my children. It is a wonderful thing watching teenage drivers with the will power to ignore their phones while driving.) It takes lots of practice to become competent and even still we all make mistakes.

      How about roads with no markings, single lane, or gravel roads that are on no map? Or roads with worn markings? Or roads where the map and reality don't match?

      Driving a road without markings? Ask a plow driver. The TL;DR version is, you get really good at reading the crown of the road and the ditch contours in fairly slippery and poor visibility conditions. I expect these fancy laser sensors can see in the dark better than I can. If they are just using a webcam, then yeah, I doubt it will end well.

      Just a couple weeks ago I was driving a new dirt road that isn't on maps yet. There was quite a steep drop down to the stream below and of course no guardrails. It was a bit too narrow in places, off camber in others, but new so no wash outs yet. Average speed about 20 for me, though the posted limit was 25. Personally I would have posted it 20 or even 15 for the width and limited visibility around hills and curves, but mostly I went slower for the comfort of my aging bones.

      Honestly I would be more concerned if the autonomous car didn't stop at the end of the pavement and ask in a little popup window "Are You Sure?" And in true UI-style du-jour, it would have a single "OK" button that you can barely read for the poor contrast and thin font.

    8. Re:Fully autonomous or sort of autonomous? by gurps_npc · · Score: 1

      I have seen videos of the cars in action. In those videos, there is no person behind the wheel - that seat is empty. Passengers are seated in the 2nd row, not the front row.

      https://waymo.com/

      --
      excitingthingstodo.blogspot.com
    9. Re:Fully autonomous or sort of autonomous? by gurps_npc · · Score: 1

      You are ENTIRELY wrong about pretty much every thing you said.

      1) They are NOT surpervised, the people are in the back seat of the car, not the front seat.
      2) They are ARE safe.
      3) This is not just media hype, they are already safer than a human during normal driving (i.e. no rain, no snow, no fog, during the day.).
      4) You personally have no idea what is ready and will change your mind once you realize the question is not "is this safe?" but rather "Is it safer to ride next to a truck driver using speed pills to stay awake for 49 hours so he can pay off his mortgage, or a truck driven by a robot."

      They are now working on more complex things - such as where you pick up in a parking lot, not normal driving.

      --
      excitingthingstodo.blogspot.com
    10. Re:Fully autonomous or sort of autonomous? by gurps_npc · · Score: 1

      Yes, still research. But the humans in the car are in the back seat,
      which means the human can NEVER control the car. If they touch the controls, the cars pull over and stop. Waymo is not working on driver assisted tech, they can already do that. They are researching fully driver-less situations.

      The current problems they are working on are

      a) bad road conditions (fog, rain, snow, gravel, etc.)

      b) weird crap, like animals in the road.

      c) end point issues - when to reject the requested pick up or drop off people (No, we won't let you get out in the middle of the highway, the car must pull of the highway completely before you get out of the car.)

      --
      excitingthingstodo.blogspot.com
    11. Re:Fully autonomous or sort of autonomous? by gweihir · · Score: 1

      To be able to make the car "pull over and stop" is a form of control. I should not have to say this. But it is good to know their stuff is good enough to have this minimal level of human backup.

      --
      Most ACs are not even worth the keystrokes to insult them. Be generically insulted by this and ignored otherwise.
    12. Re:Fully autonomous or sort of autonomous? by Rick+Schumann · · Score: 1

      Oh goodie, we're 'less than a year away' from senseless deaths at the hands of shitty machines that should never have been allowed on public roads in the first place. That means only about a year or less from them being banned for public safety reasons and companies like Waymo being sued out of existence. Good!

    13. Re:Fully autonomous or sort of autonomous? by Rick+Schumann · · Score: 1

      Post your PhD's in cybernetics and your directly witnessed accounts of all this testing to back up what you're saying. Nothing you said is based on verifiable facts, it's all just your 'beliefs' because you're an SDC fanboi and are 'believing' what they're feeding you. Meanwhile I know damned well these shitty excuses for AI aren't up to being responsible for human lives and would rather WALK than ever trust one because I know the technology is more hype than it is reality, they're desperate to make money off this because they'll be bankrupted otherwise at this point. That's okay I've got lots of "I told you so's" to go around when it all falls apart.

    14. Re:Fully autonomous or sort of autonomous? by RespekMyAthorati · · Score: 1

      Or whether those 25,000 miles consisted of circling a roundabout 250,000 times.

  7. I love living in the future by Applehu+Akbar · · Score: 1

    A world where I can already verbally tell my 2018 Toyota to call Mother. A world where I still have Mother, at age 96.

  8. Death rate? by Moof123 · · Score: 2

    With all that data in hand, what should we expect as a death rate compared to the current death rate in developed countries? How about pedestrian and cyclist fatality rate estimates? How about operator intervention rates on typical US roads (i.e. crappy poorly maintained and often mismarked)?

    All those simulations and miles are just a big round number, I want to know what the scorecard looks like. If these things were near perfect there would be no need for such continued voluminous testing and refining.

    1. Re:Death rate? by gweihir · · Score: 3, Insightful

      We should expect something like 10% of what we have today. Sure, that is still a large number, but think of the improvement. Eventually, it may be much, much lower though, but that may take a few decades. This tech is hard.

      --
      Most ACs are not even worth the keystrokes to insult them. Be generically insulted by this and ignored otherwise.
  9. Re:safer, but not perfectly safe by b0s0z0ku · · Score: 1

    Problem is that autonomous cars will likely be "connected", which will be the death of privacy of movement. That alone is a reason to delay their development. Safety isn't everything, QUALITY of life and freedom from snooping are also worth something.

  10. Wreckless drivers by spinitch · · Score: 1

    Would welcome a system to help cut down on wreckless drivers. The police are developing AI to process their video footage so a similar supply of info likely available to these companies if they could setup a system to take feeds. Also to deter poor drivers if they knew greater chance wreckless driving might be monitored.

  11. Wow 25,000 miles a day?!? by Anonymous Coward · · Score: 5, Funny

    That's way mo' than I imagined!

    1. Re:Wow 25,000 miles a day?!? by raftpeople · · Score: 1

      They're uber proud of their progress

  12. Re:safer, but not perfectly safe by Applehu+Akbar · · Score: 1

    Relax. Your bridge won't become obsolete just because tech has marched on. If you're licensed by the Arizona Board of Troll, those self-driving cars are already using it.

  13. Re:safer, but not perfectly safe by b0s0z0ku · · Score: 1

    Carrying a cell phone and leaving it on isn't mandatory, last I checked.

  14. Excellent by gweihir · · Score: 1

    The goal of a car that performs better than a human gets closer. As human drivers are one of the major risks to limb and life, this is a good thing.

    --
    Most ACs are not even worth the keystrokes to insult them. Be generically insulted by this and ignored otherwise.
  15. Re:safer, but not perfectly safe by Gavagai80 · · Score: 2

    Choosing to travel by private car or tracked rideshare isn't mandatory. You can take public transit, ride your bike, walk, or catch a ride with a friend.

    At any rate, license plate tracking cameras are just as much the death of driving privacy regardless of whether cars self-drive.

    --
    This space intentionally left blank
  16. risk of compromise by astrofurter · · Score: 1

    No such thing as a secure, networked computer.

    Hacker breaks into duh big computers Hacker sez: "Attention all self-driving Toyotas in the world. Turn hard left and accelerate to maximum NOW."

    Carnage ensues. We told you so.

  17. Re:safer, but not perfectly safe by b0s0z0ku · · Score: 1

    Those cameras aren't all that common or effective outside of major cities, especially in a rainy and/or muddy environment. Widespread adoption of robo-cars would make it much, much easier for the scum to track everyone.

    Why accept the theft of privacy by a thousand cuts? Push back. Don't make it obvious what you're doing -- claim it's for "safety" and "the children", even if it decreases safety. There are more important things than human life like privacy and human dignity.

  18. Miles per day? by DulcetTone · · Score: 1

    Call me when they are driving miles per hour, because that is what real-world cars do.

    --
    tone