Skip to content

Raw data: The safety record of Waymo driverless cars

According to a new report commissioned by Waymo and conducted by SwissRe, Waymo driverless cars are about ten times safer than cars driven by people:

The report is based on 25.3 million miles driven by Waymo cars in four US cities (San Francisco, Phoenix, Los Angeles, and Austin). This was before Waymo began freeway driving, which may affect the results. Check back next year for more on that.

107 thoughts on “Raw data: The safety record of Waymo driverless cars

  1. Justin

    Bullshit marketing. Don’t fall for it! I live in a part of the country where this just is never going to be a thing… not in my lifetime.

    Where are these cars going? What are their passengers doing? Just bullshit. Today, I am on vacation. I drove to the grocery store. I drove to my health club. I drove to a florist shop. It snowed last night. No visible lines on the road for my first trip, better later though the parking lots were slushy. Did these driverless cars park too?

    Bullshit marketing.

    1. SwamiRedux

      I'm impressed by your ability to read an internalize a study, then write a thoughtful and considered response. All within 31 minutes.

      1. Justin

        It’s a niche application. Not ever going to be widely used. A solution in search of a problem. It’s been bullshit for years. Nothing new here.

        1. Jasper_in_Boston

          Many people would prefer not to drive or are unable to do so—or simply don't want to own a vehicle. Driverless technology consequently has an utterly gigantic natural market, in that, if it can be perfected, it should radically reduce the cost of getting a ride.

          These services are only going to grow more advanced, more reliable, and more widespread.

        2. HokieAnnie

          The aging of America means there will a ton of folks in need of rides to get to appointments, visits etc but can no longer safely drive. I think you underestimate the demand.

          1. Salamander

            For that matter, would you rather send your grade school kids off on the public bus or subway system alone -- or pop them into your self driving car with instructions on where to safely deposit them?

    2. lawnorder

      I'm impressed by the fact that you either expect to die very soon or believe you can accurately foretell the future.

      In extremely snowy conditions, with which I am thoroughly familiar, even experienced human drivers have trouble telling where the edge of the road is. Driverless cars, on the other hand, can be fitted with radar, sonar, infra-red, and ultraviolet sensors, and in principle can "see" through the snow. Ultimately, they should be able to handle heavy snow better than humans.

      1. Austin

        If driverless cars can operate perfectly fine in the snow, or at least better than humans do in the snow, why do none pf the driverless companies have large presences in snowier cities? Their absence from cities with heavy snowfalls is very suggestive of their owners’ lack of faith that they will actually handle snow well.

        This feels a lot like Return To Office mandates. “Our employees say they miss in-person collaboration. Therefore we’re going to require them to return to the office full time. The fact that nobody was stopping them from going to the office every day from summer/fall 2020 to now and so offices were empty most of the time is in no way indicative of employees truly preferring to work from home.”

        1. geordie

          Both Waymo and Cruise are headquartered in silicon valley. Of course they chose cities near there to do their testing. Also both Arizona and California were willing to give them permission to test in their states without a lot of rules.

          That isn't to say having fairly predictable weather wasn't helpful in the beginning of the training but of the problems that need to be solved winter weather is likely one of the easier ones to achieve feature parity with a human. The really, reallly hard problem is predicting the behavior of other motorists. Human are incredibly good at that.

      2. ScentOfViolets

        Heh. My daughter learned a while back that a) alway have her AAA card on her - especially when it's snowing, and b) never pass on the right - especially when it's snowing. If this bit of wisdom is already built into Waymo's current driverless fleet, they're already lightyears ahead of more than a few human drivers out there.

      1. lawnorder

        Somebody's got to make the tracks. I live in a lightly populated area, and the usually light traffic becomes even lighter during snowstorms. It's common in heavy blowing snow to find yourself headed down the highway with no sign that any other vehicle has ever driven along that highway.

    1. J. Frank Parnell

      Waymo uses lidar sensors (laser equivalent of radar). Elon proclaimed human drivers got by just fine with eyes (passive visible wavelength sensors), therefore self driving cars could too.

      1. Crissa

        Lidar is expensive and not yet possible to label its data reliably.

        The occupancy system Tesla uses is 'good enough' to know where objects are for most driving tasks.

        Both types of sensors have blindspots. Lidar is just a sensor, it's not magical.

        1. NotCynicalEnough

          It can't be all that expensive as the coming Volvo EX90 has cameras, radar, and LiDar and the list price is similar to the cybertruck.

          1. Crissa

            Weirdly a Volvo xe90 carries less than half the load of a Cybertruck.

            Why would you compare the two?

            And no, it doesn't have the lidar a Waymo does.

            1. NotCynicalEnough

              That should have been EX90 and If it doesn't have Lidar than Volvo is committing false advertising as I quote from their website
              "Lidar and Safe Space Technology

              The advanced standard exterior sensor set with radars, cameras and ultrasonic sensors is in EX90 complemented with a lidar that continuously scan the environment in front of the car, enabling an accurate and high-precision estimate of the shape and size of objects in three-dimensional understanding of the surroundings."

              As for the price comparison, I don't know what load has to do with it but fine, compare it to the Tesla Model S instead. Again, the EX90 is roughly the same price with LiDar, Radar, ultrasound and cameras.

        2. J. Frank Parnell

          Yes, both types of sensors have blind spots, but the advantage of using more than one type of sensor is that the blind spots are unlikely to overlap. Safety analysts use the analogy of stacking layers of Swiss cheese. While a single layer will have several holes, the more layers one stacks the less the chance the holes will overlap.

          1. Crissa

            Those safety analysts are stupid. The problem is, the AI can't know which are the holes and the cheese and when they disagree you get a failure situation instead of added safety.

        3. rick_jones

          The occupancy system Tesla uses is 'good enough' to know where objects are for most driving tasks.

          I trust that phrasing was not intended to instill confidence...

          1. Crissa

            Because the occupancy network can tell where things are, plus or minus a couple inches, whereas a lidar could be down to tenths of an inch. But you don't need tenths of an inch to drive, you need reliability, and vision has that over lidar.

    2. Crissa

      FSD (supervised) does better than unaugmented Humans, as well.

      It is definitely not in the class of Waymo yet - but they're also trying more input at once.

      1. Austin

        FSD (supervised) is the worst of all worlds for long term driving. The person in the driver seat doesn’t need to drive most of the time, but does need to remain alert in case they need to suddenly take over. Most actual humans cannot maintain full alertness for long periods of time while also doing nothing substantial for long periods of time - eventually they get lulled into not paying attention, falling asleep, whatever. And then the crashes will be of the “car went full speed into a semi truck or home” type - there may be a lot fewer of them but the ones that do happen will be horribly catastrophic for the occupants of the car, occupants of the truck/home and even bystanders within fallout range of flying car and truck/home bits.

    1. lawnorder

      In the course of time as fully autonomous vehicles become common, legislatures will write legal codes answering all the questions about liability for accidents. At the moment, an accident involving an autonomous vehicle is covered, if at all, under product liability laws.

      1. Austin

        Nice hand waving there. As if there aren’t states that currently ignore problems - such as all the abortion bans that result in women bleeding out on ER floors and IVF facilities operating in legal grey zones now - that might also just ignore liability problems with new tech.

    2. Laertes

      This isn't the puzzler that some people think it is. The world is already full of machines that sometimes hurt people. We have a well-established body of workplace safety / product liability / other law that handles it.

      1. Ogemaniac

        The puzzle is more in that our normal auto insurance is allowed to pay out a tiny fraction of the damages in wrongful injury or death claims.

        If a Waymo splats someone, it’s going to cost them tens of millions. If you splat someone, their family will get a few hundred thousand. It’s the latter that is wrong.

      2. Jasper_in_Boston

        The world is already full of machines that sometimes hurt people.

        More basically, the world is full of people (drivers, pilots, machinery operators, machinery installers, software developers, restauranteurs, and so on) who sometimes hurt people. If a machine is statistically much safer than a human, it's much safer than a human. End of story.

        I totally get the squeamishness many people experience at the thought of being driven by a robot. I personally don't relish the though of hurtling down the freeway at 65 MPH (much less being piloted across the Pacific) by a machine.

        But as people (including me) gradually get over their irrational fear and skepticism, automated transport systems will grow increasingly common, before too long to the point of ubiquity. And at some point (freedumb lovin' America will no doubt be a late adopter), allowing people to drive themselves will be curbed, based on the overwhelming statistical superiority of machines, from a safety standpoint.

        (I'm basically with Drum on this, though I think the timeline for the arrival of this brave new world is a bit slower than he deems: I reckon human driving will well and truly start to collapse in another 35 years or so.)

        1. Crissa

          That's not how that works.

          A pilot or person operating is not liable for a mechanical failure outside their ability to sense or predict it.

          In those cases, the liability goes to the manufacturer of that machine.

          1. Austin

            And everybody in the future will maintain their self driving cars really well, unlike today in which it’s not uncommon to see cars held together with duct tape.

            1. lawnorder

              We already have on-board diagnostics. If desired, or legally required, self-driving cars should be able to monitor their own mechanical condition and need for service, and refuse to operate if dangerously neglected.

          2. Jasper_in_Boston

            A pilot or person operating is not liable for a mechanical failure outside their ability to sense or predict it.

            Huh? Where did I write anything about "liability"?

              1. Jasper_in_Boston

                Why would you bring in the concept of liability to a comment that had nothing to do with that word? Seemed non-responsive, is all I'm trying to say.

        2. rick_jones

          I personally don't relish the though of hurtling down the freeway at 65 MPH (much less being piloted across the Pacific) by a machine.

          But you are being piloted across the Pacific by a machine. By the time the aircraft reaches cruising altitude, if not earlier, the autopilot is flying the aircraft. And that goes back decades at this point.

          1. Jasper_in_Boston

            But you are being piloted across the Pacific by a machine.

            You mean I'm mistaken in believing that jets are required to have human pilots?

            Right. Didn't think so.

            1. lawnorder

              The fact that a jet has a human pilot in the cockpit does not in any way obviate the fact that when the autopilot is on the plane's being piloted by a machine.

              Modern autopilots have the capability to go from runway threshold to landed and stopped without human interference. The only real NEED for human pilots is when taxiing.

              1. Jasper_in_Boston

                the fact that when the autopilot is on the plane's being piloted by a machine.

                Agreed. I'm not claiming otherwise. Nonetheless, there's a considerable difference between getting on a plane with human pilots who have various automation technologies at their disposal, and getting on a plane without human pilots.

                I'd guess when human passengers are involved, the latter doesn't come to pass for many decades, if ever (even though it won't be necessary from a safety standpoint) because of the fear factor.

          2. bouncing_b

            I wonder if this has clues to resolve what Austin raised above

            Most actual humans cannot maintain full alertness for long periods of time while also doing nothing substantial for long periods of time - eventually they get lulled into not paying attention

            How do long-distance over water pilots stay focused?

            Side story. I used to fly across the Pacific regularly for work. Before 9/11, I often used to ask to sit in the cockpit, and many times was welcomed there. My impression was that the pilots were bored shitless, and were happy to have a new face if I could make entertaining conversation.

            1. lawnorder

              There is compelling evidence that more often than airlines like to admit pilots nap while the autopilot is flying the plane.

  2. jamesepowell

    I've see a lot more Waymo cars around Los Angeles in the last month, including on the freeways. Don't know who is using them or for what.

    1. Dana Decker

      In Los Angeles, Waymo taxi service is now happening. I've seen them many times and very close up and am impressed with the way they work. They tend to be cautious, like not nosing into an intersection to make a (legal) right turn on red. I'm convinced Waymo (not Musk's whatever-he's-touting) will shake up the transportation business and eliminate thousands of jobs.

      1. shapeofsociety

        It'll make transportation safer and more accessible, and free up thousands of workers to do something else of value, thus growing the economy. Massive win.

      2. Austin

        The minute the streets are full of “perfect” self driving cars, I’m just going to start cutting across the street wherever and whenever I feel like. After all, why should I care about crossing at a marked crosswalk with the light? If the technology is “perfected” the cars will stop and it’s Somebody Else’s Problem if gridlock results from everyone jaywalking and the self driving cars are too afraid to do more than inch forward constantly on their journey across, say, DC or San Francisco or Manhattan.

        1. Salamander

          And teenage boys will delight in throwing themselves in front of self driving vehicles moving at top speed, because it's what they do. What could go wrong?

        2. lawnorder

          Self driving cars are still subject to the laws of physics. An autonomous vehicle's stopping distance can be expected to be a bit shorter than a human driven vehicle's stopping distance because computers have shorter reaction times. However, if you step out in front of an autonomous vehicle inside its stopping distance, it will hit you, just like a human driven vehicle would.

          Also, human drivers will generally try to avoid hitting pedestrians, even jaywalkers, so self-driving vehicles won't be much different.

    1. bouncing_b

      No, "better than human drivers" is the appropriate bar.

      First of all, better is better. Already I'm less likely to be injured by a Waymo car than one driven by meat.

      But mainly, self-driving tech improves by experience - and benefits from the collective experience of thousands of its "colleagues". It can be expected to continually improve.

      Unlike us humans who learn once, when we're young, and then only pick up bits of new knowledge and can be expected to get only slightly better.

      1. Crissa

        We also can learn from other's mistakes; and collectively we can change training data and then the tools and situations in which we're applying those skills. So each generation is in fact better.

        But those also apply to AI, which means it will eventually surpass us.

  3. shapeofsociety

    For people who cannot drive, this is going to be a wonderful game changer. People who can drive take it for granted and never stop to think about how hard it makes a person's life to not be able to drive. Having to either pay through the nose for an Uber, figure out public transportation scheduling, or work out a ride with friends or family every single time you want to go anywhere is a huge burden. Go Waymo!

    1. Jasper_in_Boston

      +1

      I mean, folks who don't drive already have it a lot better than they did 30 years ago, when their only option was the predictably shitty local taxi monopoly.

      But yeah, when you add ultra-low-labor costs to the concept of app-based rides on demand, car-free living becomes more and more plausible for millions of people. Being liberated from the tyranny of mandatory car ownership would be gigantic win for society.

      1. Austin

        It’s too bad people couldn’t just be freed from the tyranny of mandatory car ownership by designing cities around walking, biking and transit. You know, like how most European and Asian cities already are, with car ownership well under 50% in all of them and - at least in Western Europe and Japan - living standards very comparable to our own.

        But that’s just unAmerican crazy talk! Next thing you know, I’ll be advocating for more walking/biking and less high fructose corn syrup and high processed food consumption to deal with higher obesity in America vs Europe or Asia! Or more gun restrictions to enjoy lower murder rates like Europe and Asia have! Freedom isn’t free, and ours is especially costly.

        1. Jasper_in_Boston

          It’s too bad people couldn’t just be freed from the tyranny of mandatory car ownership by designing cities around walking, biking and transit.

          Even the very most car-free enabled cities face the last mile problem. I live in a city with the world's largest single subway system both by trackage and ridership, but even here there are millions of people who live beyond walking distance of a subway station.

          I don't disagree about the desirability of designing our urban spaces to enable car-free living. And moving forward, we should keep this in mind. But at this point, in America at least, retrofitting our our already designed and built urban and suburban spaces toward this end would surely be astronomically more expensive than simply letting the market continue to advance the frontier of app-based rides married to driverless technology.

        2. Crissa

          We can do both. As more is automated, annoying road features (but which are safer) like protected bike lanes and roundabouts, will be annoying a shrinking pool of drivers - and hence, a shrinking pool of voters.

    2. bouncing_b

      The real payoff will be when you can app a self-drive car and don't need a personal one parked in front of your house. Think of how our streets will change when they're not lined with parked cars! Plenty of space for bike lanes, for one thing.

  4. skeptic

    I am excited about what appears to an approaching mobility revolution! With a fully AI controlled transport system, insurance rates might be next to nothing and the need to build tank like cars to protect one against other drivers would end. We might see golf cart like vehicles in the future with minimal to no crash protection at very very reasonable prices. With such vehicles multiple new lanes of traffic might become available. There are many exciting features to watch evolve in transport.

  5. middleoftheroaddem

    I have been in a Waymo twice (San Francisco). Safe, easy and no complaints.

    I wonder, at least its possible, if all Waymo's COULD suffer from the same type of failure: meaning, a strange circumstance happens, lets say broad flooding in a particular city, that the software does not know how to properly handle or the software has a bad solution. Possibly, multiple Waymo's would make the same error.

    Note, I HOPE that Waymo has sufficient human overrides, to avoid 'group' error.

    1. James B. Shearer

      "I wonder, at least its possible, if all Waymo's COULD suffer from the same type of failure: ..."

      Sure it is possible, they could all go berserk trying to handle a time change or on February 29th or something. Perhaps more concerning would be a security problem allowing a malicious actor to wreck havoc.

  6. J. Frank Parnell

    Was just in downtown San Francisco over the weekend. Driving was awful, narrow streets, grid locked traffic. Waymo taxies were ubiquitous and seemed to doing okay.

    1. iamr4man

      My daughter, who lives in SF, uses Waymo frequently and prefers it. She said it felt weird at first but she got over that after a few trips. For me, it is kind of weird to see in person. I lived in The City for 17 years and it is a place with narrow and confusing streets and HUA pedestrians. Waymo seems to be doing very well there, but of course every hiccup gets a lot of publicity.

      1. Jasper_in_Boston

        I hate that I need to make this observation, but make it I do: driverless technology is a big safety enhancement in a society where violence against women is tragically too common. Getting into a car with a stranger just seems fundamentally risky to me.

        1. lawnorder

          Is it common for taxi, Uber or Lyft drivers to sexually assault their passengers? If it were a significant problem , I would have thought I would have heard about it.

          1. Jasper_in_Boston

            I'm not a ride-hail basher. Far from it. Mobility for the carless has been vastly improved in the US over what it was 30 years ago.

            That said, sexual violence against women riding Uber/Lyft seems to be a very real thing. I don't think it's a phenomenon that justifies a panic, mind you. But it bears watching (and these firms should continue to be pressured to improve their practices in this area).

            https://helpingsurvivors.org/rideshare-sexual-assault/statistics/#1

            And needless to say, driverless cars by definition have to safer with respect to this particular risk.

      2. Batchman

        iamr4man, I don't mean to cause Unrest, but what do you mean by "HUA pedestrians"? I can find only two meanings: one is a military acronym for "Heard, Understood, Acknowledged" and the other suggests you are describing persons of Chinese descent (admittedly a large presence in an otherwise Western Culture city). If there's a third meaning that I missed, I'm In Praise Of Learning what it is.

    2. Crissa

      The traffic regularly locks up for a couple hours, it's not a big deal - that's why we have sidewalks, lane sharing for bikes, and subway light rail.

        1. rick_jones

          They are certainly the most convenient. There are few places to get a car into the "subway light rail" - aka Muni trains ...

          1. Crissa

            Well, I was trying to lump the Muni in with BART there.

            Personally, I try not to drive across town during those times, and park as far out as possible to change to transit.

  7. skeptic

    Once the entire transport system converts to AI control then there would likely be massive improvements in safety. As it is now the system is handicapped by the very non-rational behavior of humans. Often when there are accidents with robocars it is a human at fault and not the AI.

    A neural net could create safety protocols that humans could not match. For example in the classic trolley problem, an AI controlled network could instantly respond in some pre-programmed or computed way and potentially avoid injury; such instant precise, coordination is not possible with humans. One would expect that in a fully optimized system personal injury might become nearly impossible.

    1. Austin

      The massive improvements for safety will likely be Las Vegas* strip style barriers forcing all pedestrians to not jaywalk, because - as I mentioned above - as soon as everyone is sure the cars won’t hit them - there’s no incentive for pedestrians to walk to a marked crosswalk and wait for a light to change. Those barriers make the streetscape look particularly ugly - which in Vegas doesn’t matter since everyone’s eyes are upwards to the lights, but might not look great in older downtowns.

      But building pedestrian barriers will be costly, so I expect only well-off municipalities to do so. The rest will be a free for all: look up old photos of cities and you’ll see pedestrians, horse and buggies, streetcars and everything else in the street, with none of them able to more quickly. That’s the future in municipalities that can’t afford to corral the pedestrians to the sidewalks, once the threat of injury that currently keeps most pedestrians out of the road is eliminated.

      *Vegas has them now because of all the public drunkenness right next to a wide roadway built to 55mph driving standards. (Yes it’s zoned for less, but absent ubiquitous speed cameras, people will drive up to the design limits, as traffic engineers know well.) The drunkenness is what causes the pedestrians to discount the threat of injury that keeps pedestrians in other cities from just stepping into traffic without looking.

      1. Crissa

        Maybe just design the street to promote slower driving rather than wishing for speed cameras that rich people will ignore and fines which poor people can't escape?

  8. jdubs

    The final hurdle to get over will be from a regulatory capture perspective. Will we change road design, traffic laws and liability enforcement to benefit and improve the overall driving experience? Or will we go the other, more well trodden path of making manual driving a more difficult experience in order to benefit the autonomous vehicles?

  9. Justin

    Who runs Waymo? Is it a trustworthy well meaning company with a track record of doing good things for humanity? Like Lambs to the slaughter… 😂

    1. rick_jones

      A "private" vehicle mostly sits in one place. A "Waymo," like a taxi and perhaps moreso, will have a much higher trip rate. Normalizing to miles driven is indeed the correct comparison.

      1. Jasper_in_Boston

        Also, how severe were those 700? What if 98% of them were fender-benders? Also, how many were the fault of Waymo as opposed to other cars?

  10. varmintito

    I love Kevin's work, but this is one fight I will never surrender:

    The study indicates AI is 90% safer, not ten times safer. 10% as many deaths, therefore 90% fewer, than human-driven.

    The only time you can express the concept of "less" or "fewer" with a multiple is when you are comparing two or more reductions (hypothetically, comparing the reduction in fatalities when you use AI, vs the reduction in fatalities when you use technology X). If AI reduces fatalities by 90% and technology X reduces them by 45%, then you can say that AI reduced fatalities two times more than technology X. AI produced twice the reduction that technology X did, and therefore is two times safer relative to the baseline.

    Human-driven created ten times more fatalities than AI, but AI did not create 10 times fewer fatalities than human-driven. It produced 90% fewer.

    There is usually a right way to describe mathematical relationships in words.

    1. Jasper_in_Boston

      I hate this usage too, But unfortunately it's become pretty ubiquitous, rather like saying a "thirteen percent" increase when what should be written is "thirteen point" increase.

  11. Owns 9 Fedoras

    One more testimonial -- I too have used Waymo in SF, exactly as I would have used Lyft. The system centers on its iOS app, which seems well-designed with a smooth UI. The app directs you to the pickup location (more clearly than the Lyft app does).

    As the car approaches, you see your own initials in the lit dome on the roof, so you know your vehicle in case there are two or more. You tap a button on the app to unlock the door; it is impossible for anyone else to jump in until you do. The car doesn't go until you tap "Begin Ride" on the dash display, thus there is no hurry to get multiple people settled, belts on etc.

    Inside the car, there's an informative display screen on the dash, which shows what the car is "seeing", i.e. it shows icons for nearby vehicles and icons for pedestrians as far as half a block ahead, updated in real time as they and the car move. Plus you can tap the dash display to select your style of elevator music to listen to.

    The cars drove conservatively, and smoothly manouvered around double-parked vehicles, made legal left and right turns and merges, etc.

  12. KJK

    Probably safer because Waymo doesn't need to mess around with it's phone, play with the car's entertainment system, adjust the HVAC, deal with crying/screaming children in the back seats, worry about being late, get frustrated with traffic, get angry with other drivers, recklessly disobey traffic rules, or act like a total schmuck behind the wheel.

    I have never been in one, but do they have an animatronic driver behind the wheel, like the "Johnny Cabs" in "Total Recall"?

      1. KJK

        Too bad, I am sure that Disney could put together a really nice animatronic driver for its entertainment value. A blow up plastic driver would be ok, like "Otto the autopilot" in "Airplane".

  13. jlredford

    In reading through the report, one interesting aspect is that they compared the Waymo accident rates to not just the overall rates for the same zip codes, but also to the rates for new-ish cars. These have lots more safety features than cars older than 2017, like collision warning and automatic braking. To be precise, a much higher proportion of new-ish cars have these features, while on older ones they existed only on the high end. Newer cars are also likely to have older, better drivers. They really do show lower accident rates than the overall population, by about 20%, but Waymo's are still a lot better.

    The human-driven rates, though, are for everybody, including drunks and morons. You know, typical drivers. I wonder how this would look compared to comparable taxi rides in either cabs or ride-shares. Those are more professional human drivers. They might still be drunk, and might still be idiots, but there are more penalties for that.

    As for the advantages of using such services instead of having to own a car - well sure. That's why taxis and ride-shares exist. The key point is how affordable it will be. Waymo has sunk tens of billions into this. There's a lot of extra hardware on every car for it, and they have to carry a lot of insurance. They can't subsidize it forever, and lots of companies have already given up on the concept.

Comments are closed.