Hacker News new | past | comments | ask | show | jobs | submit login
Waymo CEO Says Alphabet Unit Plans to Launch Driverless Car Service (marketwatch.com)
376 points by sidhanthp 4 months ago | hide | past | web | favorite | 348 comments

I'm excited. If you have been alive a substantial amount of time you are easily running a lifetime average of about 1-2% chance of dying in a car accident in your lifetime. Things have gotten much safer the past couple decades so that number is somewhere around 0.75% or lower now. I want to say being injured in a car accident is somewhere around 20-30% lifetime chance. Cars are incredibly dangerous and this will save a ton of lives.

I ride a motorcycle every day as my primary way of commuting to work. The only time in 10+ years I've been in a serious accident is when inattentive car driver rear-ended me on a highway. In his own words: "but I didn't see you". In the bright daylight. I was wearing hi-viz jacket.

So I am super excited about any substantial progress in self-driving cars deployment.

Fun fact, you're ~25 times more likely to die on a motorcycle mile for mile than a car.

Stay safe out there.

Every motorcyclist knows this, yet, driving a motorcycle is so awesome.

The danger are other cars on the road. I've been thinking about creating computer vision tech to spot inattentive drivers, it's something I think motorcyclists could really benefit from (myself included)

I've known 3 cyclists killed or severely injured by road debris. At freeway speeds and density there may be no way to avoid going down. It may still be a larger moving vehicle that finishes the job, but once you're on the ground you're lucky to walk away.

I think with AR more common we'll all be able to 'tag' other drivers. So drivers who have a bad reputation will be flagged.

I imagine a system in which 'poor driver' tags would decay with time. Data is shared.

Would be neat for self driving cars to update this data in real time, so you could see a data overlay of their decisions and intentions. A long green arrow would show their intended path, icons show you if they have seen you, timers counting down e.g. I'm pulling away in 3, 2, 1 ... now

AR is going to be fun.

I'm not so sure it'll be fun for everyone... that is only a couple of steps away from a Black Mirror episode (specifically S03E01, Nosedive). The risk of 'poor driver' ostracism/stigma is significant.

The road is a public space where privacy is not expected, and your behavior directly affects the safety of others. I'm not sure that we ought to protect "safe space to be a bad driver" the way other personality differences ought to be protected.

You assume that bad driving is a fixed trait. In addition, driver 'ratings' are given by other humans in the proposed system; humans are biased and sometimes even spiteful on the roads. A rating system sounds great in theory but wouldn't actually work.

Similar tech has been done with cameras places in cars, but from a motorbike to monitor cars nearby would not be so easy, if the goal is to monitor the drivers face position (light reflection on glass is a major hurdle, i think). Perhaps it is possible to get the information based on car movement.

I'm afraid that while driving motorcycle it's best to keep your eyes on the road as much as possible. Any system wanting to notify you about nearby inattentive driver will distract your attention from actual actions of that inattentive driver.

I was thinking about a sound, like bbzzzzzttt to give you a heads up just those 100ms earlier than your reactions would've given you, I really think those milliseconds count in these kind of situations

> The danger are other cars on the road

And potholes, debris, slick spots, ... the kinds of things that might inconvenience or damage a 4-wheeled vehicle can be deadly to a 2-wheeled vehicle at 60+ mph. Losing control of a car, you might skid, spin, or flip. Rolling a car can be deadly. Losing control of a motorcycle at high speed will almost certainly cause serious injury or death.

You're probably not as good a driver as you think you are. After all, we can't all be above average.

Sorry, I'm from the Netherlands and most of the road are really good quality compared to, say, the roads in the US and Belgium. The moment you cross the border you know you're in Belgium because of the bad road quality.

I'm glad you have good roads. The last part still stands -- Lake Woebegone effect.

> The danger are other cars on the road.

Both my accidents were with HGVs[1], the reality is the danger is everywhere

[1] Only one was their fault, the other entirely mine

I wonder how it breaks down for motorcycles. Some part of the danger is because you're vulnerable and you're riding among big metal objects (cars). Another part of the stats would be explained by dangerous riding (speeding and such). The other part would be that motorcycles are just quite dangerous by themselves given that there's no crumple zones or shielding when you're on a bike. Autonomous vehicles should eventually result in cyclists and motorcyclists being safer, but perhaps not completely.

If you read the report, you'll find motorcyclists are more likely than car drivers to speed, drive drunk, and do both at the same time. And they frequently don't wear helmets.

There's no way to make motorcycling as safe as driving, but last time I did the math I found you can recover almost an entire order of magnitude just by obeying the law and wearing a full face helmet. And there's plenty more you can do from there -- take an MSF course, avoid congested areas during your first year on two wheels, and get a modern bike with ABS. Statistically speaking, the benefits of all these things are compounding.

That can only get you close to the ‘average’ car driver. The equivalent cautious driver is significantly ahead of that mark.

PS: 72.34 deaths per 100,000 per year means the average lifetime motorcycle rider has a ~4% chance of a fatal accident. Which is approximately the same death risk of a first time suicide attempt. Regular riders are even worse off.

Definitely! I don't mean to downplay the inherent dangers of motorcycling. My point is simply that motorcyclists, as a whole, skew less cautious than drivers do. So e.g. a rider who never drinks and drives makes up more statistical safety ground than a driver who never drinks and drives.

I suspect that in the US, the lack of caution in bike culture is exacerbated by how relaxed our laws and regulations are (I've looked for data on this, but I haven't found great cross-country comparisons). We just don't impose on bikers in the same way we do on drivers. For example, we mandated ABS on all new cars nearly two decades ago, but we still don't require it on new bikes.

In fact, even in the US, I suspect riders have European laws to thank for the safety features on their bikes. It can't be a coincidence that, for example, Zero Motorcycles began installing ABS with their model year 2015 bikes -- just in time for the EU law that went into effect in 2016.

They also drive a lot less. They are 0.7 percent of all miles driven in the US, but 14 percent of all fatal accidents.

You personally can decrease the risks, but not enough to adjust for the 35x per mile risk and make them safe to daily drive.

According to my recent MSF course somewhere around 60% of motorcycle fatalities and serious injuries are self inflicted. Some 40% involve more than 0 alcohol.

Nothing made me feel safer on my new motorcycle than learning that stat. Don’t do dumb shit and your personal danger per mile can be quite okay.

Also you’re most likely to die in your first and your third year. Then the probability drops off a cliff.

First and third? What about the second

Likely the same effect as airplane pilots, where the 100th through 250th flight hours of your career are the most dangerous.

Stage 1: unskilled, unaware => die

Stage 2: unskilled, aware => not die

Stage 3: unskilled, confident => die

Stage 4: skilled => live

Difference is that certificated flight instructors won't let you solo until you're well into Stage 2. But any noob squid with a credit card can ride off a lot with a liter-class bike, no questions asked.

Does this to private pilots?

I think you meant to include "apply" somewhere in there. If so, then yes. I don't know anything about military flight training, but except for that I believe every pilot starts out as a private pilot.

Lower chance than first and third. First year you’re a newb and likely to make mistakes. Third year you get confident and sloppy and likely to make mistakes.

Most common way to die is running off the road when going too fast into a corner. Followed by being too aggressive in intersections and failing to avoid other people making mistakes or not seeing you.

If a car forces right of way and you could’ve stopped but didn’t, it’s your fault. Even tho legally it’s their fault but you’re the one with less crumple zones so you gotta drive defensively.

I thought it was mileage not years. First 2k miles are second most dangerous. Next 2k miles are most dangerous because you get comfortable enough to make mistakes.

Before I went for a motorbike licence I research some of these stats.

- A decent proportion of accidents are on unregistered bikes or riding while drinking.

- Speeding, as always

- A big risk factor is older people returning to motorbike riding as they remember their old skills but dont have them any more.

- Having ABS is something like a 30% accident reduction.

There's still a bunch of risk on a bike but for someone sensible you can improve your odds a fair way. Though I guess you could argue that personality would likely have well reduced odd than average in a car too so its still a relative increase...

When I started riding, I noticed the same thing. The HURT report shows unlicensed/uninsured young people (I was 28 when I got my license) as being so massively overrepresented in motorcycle injuries that it was mind boggling.

Sad, but it also means more organ donors.

Simply put - you are exposed.

Anything (accident) happens you fall and hit something - ground or otherwise, or are run over at times, again because you are exposed. You lost balance? Same. You just got on sand at even half the normal speed? You might still fall and hit. Wet surface, mud, speak breaks, same same. You hit another bike? Same (might be for both). You hit car? Same for you, but not for the car driver.

When the bike falls you are the one who takes the hit. Most of this doesn't happen in a car. Just my observations as a biker. Yes, riding at sane speeds reduces these risks a lot.

Stability is another huge concern for motorcycles. If you lose traction on just one of your wheels, you could crash. While losing traction on 2 or maybe even 3 wheels in a car is much easier to recover.

I also remember being told statistics when I did the MSF training course many years ago that a very high percentage of single vehicle motorcycle fatalities involved substance use. It was really really high - 80+% or something.

I rode motorcycle for 8 years as my only mode of transportation and can clearly remember how little alcohol or marijuana was needed to notice a significant performance degradation (much more noticeable effects from small amount of substance use than when driving a car).

Yeah, from what I read when I first started riding, there are a few really easy things that can significantly reduce your risk:

* Wear a goddamn helmet

* Don't drink and ride

* Don't speed

* Get a bike with ABS

> Fun fact, you're ~25 times more likely to die on a motorcycle mile for mile than a car.

According to recent US numbers (Uber's Elevate study), it's about 38x (per pax mile).


EDIT to add: table on page 17.

But most of that is in the age range of 18-24. 90% is in city, and of that 70% is turning left at an intersection.

Which mile is matters, as is who is riding the bike.

Dang. If that's true, then combining that with the parent comment's statistic (1-2% chance of dying in a car accident over a lifetime), then you have a 25-50% chance of dying in a motorcycle (assuming you drive the same number of miles on your motorcycle over your lifetime as the average person drives in a car). 1-2% seems way too high.

According to Forbes [1] it’s 1/88 for any kind of vehicle accident. 1/303 specifically in a car crash, which is definitely a lot lower than the original 1-2% quoted (though I’m sure it varies depending on where you live). Also, I guess even bikers don’t ride 100% of the time in place of being in / around cars.

[1] https://www.forbes.com/sites/tombarlow/2011/05/10/the-odds-o...

The silly joke here (at least in Italy) is to say: go 25 times faster than a car. (of course ignoring that it's still 25 times more likely)

But motorcyclists are on average get in lesser amount of serious accident.

When you know that how dangerous is careless driving, you drive more carefully

And horses are 20 times more dangerous than motorcycles.

I love motorcycles. I was out on some back country roads in July visiting my parents. It was a beautiful day, when an inattentive 93 year old man decided to make a left turn into oncoming traffic. I was on a skinny 2 way double yellow road and his truck had a trailer, he blocked both sides. I broke every bone in my right leg and was lucky to not have it amputated. I was wearing racing gauntlets, helmet, jacket, armored jeans and riding boots so I was relatively uninjured. The hospital staff told me typical motorcycle accident patients aren't even responsive when bought in.

Not sure if I will ride again. I love it but despite my gear, practice, having taken advanced riding courses, it really creeps me out that I got into a situation where I could only minimize my injuries.

Unfortunately, I only see the issue of inattentive elderly drivers getting much worse in the coming decades as the boomers age. There is not an effective mechanism in place to ensure drivers with limited sight, vision, reaction time, and mobility are actually still capable of passing a drivers' test.

I see this when I'm cycling, years ago the problem was agitated young men, now it's most definitely inattentive old people.

> In his own words: "but I didn't see you"

Ah, the infamous SMIDSY. "Sorry, mate, I didn't see you!" has claimed many of our peers.

'self' driving cars are going to have a _even harder_ time at this task... and they are not going to apologize. The assumption that pre-programmed computers on wheels are going to compete with wetware in the real world is wrong. Instead of admitting it, the proponents will treat the people as bugs and demand more rules to make the programming easier.

fatality. broke the rules. NOTABUG.

That doesn't make any sense to me. There are many challenges in fully autonomous driving, but detecting objects around the car is one of the easiest to overcome.

We have radar and lidar, tried and tested technologies being used for decades. Unlike the wetware you speak of, this system doesn't get tired or distracted.

Yes motorcycles are smaller than cars, but that won't be a problem for radar/lidar, and with proper sensor placement, there won't be any blind spots.

> detecting objects around the car is one of the easiest to overcome.

Not sure how you can say that, given recent accidents (crash into pedestrian pushing a bike across a street, crash into a stationary lane divider). Possibly other challenges are even harder, but describing a task that real world driverless cars repeatedly fail in as "the easiest" strikes me as overly cheery/Pollyannaish.

Can you provide sources for the accidents you are referencing? I'd like to get more detail about what happened and whose technology was at fault.

It boils down to execution and application of the technology. The key elements, radar and lidar make it really easy to detect objects around the vehicle in real time and react quickly. Now how an autonomous vehicle company decides to use these technologies matters in the overall effectiveness. For example, Tesla doesn't use the lidar at all for cost savings. Tesla relies on radar and machine vision for localization which to someone like me is scary. On the other hand, waymo uses lidar. Look at their accident rates, very impressive indeed.


Specifically, the ones I had in mind are:

1. Uber crash, Tempe, Arizona, 2018-03-18

2. Tesla crash, Mountain View, 2018-03-23

> Autopilot may not detect stationary vehicles at highway speeds and it cannot detect some objects.

To be more precise, then: I agree with you that might be "very easy" to detect objects and then, say, stop. But apparently, self-driving cars would then stop all the time needlessly, which is why they're not programmed that way.

So, what is hard, then, it appears, is to avoid crashing into people or objects, but still keep driving when safe.

[1] https://en.wikipedia.org/wiki/Death_of_Elaine_Herzberg

[2] https://en.wikipedia.org/wiki/Tesla_Autopilot#Mountain_View,...

Yup, nothing can ever work. Strong position.

If nothing means semiconductor based binary Turing machines; yes.

I expect a pivot (>10y) to wetware if people really do* demand to give up their dv/dt. Rat neurons were flying F-22 sims a decade ago.

*(bad idea, I don't think we will)

So were auto-pilots.

Autopilot works only in a highly constrained environment.

I think this is one of the most overlooked safety features of a self driving car. It is not just safe for the car occupants but it will be safer for pedestrians and 2-wheeler riders - motorcyclists and bicyclists. There was a video of a Tesla just following a cyclist at a safe distance in a single lane road. The car was in a way protecting the cyclist from other cars. I will not be surprised if the motorcycle deaths come down dramatically once self driving cars become more pervasive on the road.


I hope "err on the side of caution" is embedded into it. Though in driving erring in caution to prevent one collision can cause another. Another motorcycle (scooter really) commuter whose biggest issue is turns in front me with w/o indicator or sharp cuts in the front.

i ride a motorcycle and will be just as terrified if not more of automated vehicles, so i don't understand your reasoning at all. the patterns of poor human drivers are actually pretty predictable. in the future, i will need to learn everchanging patterns with automated driving systems.

Even if he were able to predict a rear end collision, how would he avoid it -- stuck in traffic or at lights?

I'm sure you'll learn the ever changing patterns ok.

that's a dismissive comment. i didn't mean it's safe now, and that situation isn't really avoided by anything other than constantly checking your mirrors, which isn't practical and maybe won't even work against that. however, i don't really understand the anecdote or reasoning because it sounded like the commenter went from "humans are bad" to "robots will be good". my point is that i view them both as dangerous but robots will be dangerous in new ways.

have any self-driving cars been shown to work in motorcycle lane-splitting areas like what happens in california in the h.o.v. lane or anywhere else in the world besides the other u.s. states?

Road safety is a regulation problem first and foremost, we could have safer roads today if we cared. Also, less cars are needed on the roads, not more and especially not driverless ones.

Yep. There is substantial evidence to suggest that better road design stricter driving standards enforcement changes lives.

There is no evidence other than exponential curve worship to believe that driverless cars on the existing road network will ever be less lethal than the surprisingly high benchmark for sober drivers, and plenty of reason to believe they will increase car use. Waymo demonstrating their cars are surprisingly competent and less aggressive than human road users is very different from demonstrating their AI is smart enough to make a fatal errors less frequently than every half billion miles in real world driving conditions.

People do care and the roads are getting safer but it's a gradual process and has limits.

I find it just insane that it is socially accepted and normal to walk besides cars, cross in front of them, etc.

They are killing machines driven by idiots with low IQs. Why do I need to put my life in their hands every single day?

They were grandfathered. It's hard to imagine how cars could be introduced into the world today as a new invention that kills a million perfectly healthy people each year.

That's the classic John Stossle argument he made in the 90s with natural gas.

I dunno. There is a huge risk with driverless cars. I'm still questioning the tech. This is an incredibly hard problem and edge cases mean people die. This is one technology I won't be an early adopter of.

I just paint it like this:

If you made a teleportation system where you could step in and come out the other end anywhere in the world, but there is a 1 in 1000 chance you step in and turn to green goo on the other end, should we still adopt it?

You basically save millions of lives, but people have seem to come to the notion that they protect their own fate when in a car (which is 100% untrue). You may be able to mitigate some accidents, but getting T-boned at an intersection when you have a green light, but no visibility, is possible every single day. Your wheel bearing could blow out on the highway and send you rolling into a ditch. A deer could step out in front of you and go through your windshield and kill you. You have no ability to save yourself other than praying.

It's people like you that make people wary of mandatory vaccines.

I very much understand your hesitation; And I agree that the tech is still unproven - also because there's no single "Tech", but different approaches from different companies with highly varying aims and safety standards.

But the edge-case argument also applies to human drivers. We suck at sudden, unexpected reactions. We panic. We overreact. We die. We kill.

I'm not saying that Self-Driving cars should just go on the road now no matter what - but the point of "edge cases kill people" is a perfectly valid argument against the status quo.

> This is one technology I won't be an early adopter of.

Oh, but you will.

> They are killing machines driven by idiots with low IQs.

You don't have to go that far. Smart people drive cars and they fuck up just as much. We (humans) just suck generally at that task, and with more and more assistive features coming in, the attention paid to driving is going down even more.

You put your life in the hands of plenty of machines, every day.

To your point, ~3 years ago I didn't die in a car accident by literally a few inches. I'm with you - car safety is so important.

I don’t die in a car accident every day by a few inches.

I’m totally serious. This is true for most every car trip. You trust yourself and the idiot in the other lane with your life every single day. It’s amazing how well cars protect us these days when we crash them. I’m excited about having them avoid the crashes.

I suspect you’re both writing about different inches. For my own example I offer an experience from my teenage years where a bus passed me with only one inch to spare from scraping that handlebar while on my other side there was an uneven slatted wooden fence with 1-1.5 inches to spare before it hit that handlebar. (Bicycle, not motor).

Yes, we are. I was referring to a really scary (and very rare) occurrence.

I understand that but was pointing out how close to doom we are very frequently in traffic. Cars avoid head on collisions by single digit feet at relative speeds of 100kph, controlled by barely-trained, tired and distracted people and this is totally normal. Get just a bit closer and it’s totally terrifying.

Single digits feet is normal. When I wrote “one inch” I wasn’t exaggerating.

What if you're in the car that the self driving algorithm decides it's safer to crash and kill you than hit whatever it thinks it was going to collide with?

Edit- downvoters please state your beliefs, there is no right or wrong answer to points 1-7 below.

[1] Imagine you’re in a self-driving car going down a road when, suddenly, the large propane tanks hauled by the truck in front of you fall out and fly in your direction. A split-second decision needs to be made, and you can't think through the outcomes and tradeoffs for every possible response. Fortunately, the smart system driving your car can run through tons of scenarios at lightning fast speed. How, then, should it determine moral priority?

Consider the following possibilities:

1. Your car should stay in its lane and absorbs the damage, thereby making it likely that you’ll die.

2. Your car should save your life by swerving into the left lane and hitting the car there, sending the passengers to their deaths—passengers known, according to their big data profiles, to have several small children.

3. Your car should save your life by swerving into the right lane and hit the car there, sending the lone passenger to her death—a passenger known, according to her big data profile, to be a scientist who is coming close to finding a cure for cancer.

4. Your car should save the lives worth the most, measured according to amount of money paid into a new form of life assurance insurance. Assume that each person in a vehicle could purchase insurance against these types of rare but inevitable accidents, and then, smart cars would prioritize based on their ability and willingness to pay.

5. Your car should save your life and embrace a neutrality principle in deciding among the means for doing so, perhaps by flipping a simulated coin and swerving to the right if heads comes up and swerving to the left if its tails.

6. Your car shouldn’t prioritize your life and should embrace a neutrality principle by randomly choosing among the three options.

7. Your car should execute whatever option most closely matches your personal value system and the moral choices you would have made if you were capable of doing so. Assume that when you first purchased your car, you took a self-driving car morality test consisting of a battery of scenarios like this one and that the results “programmed” your vehicle.

There’s no value-free way to determine what the autonomous car should do. The choice presented by options 1–7 shouldn’t be seen as a computational problem that can be “solved” by big data, sophisticated algorithms, machine learning, or any form of artificial intelligence. These tools can help evaluate and execute options, but ultimately, someone—some human beings—must choose and have their values baked into the software.

Who should get decision-making power? Should it be politicians? The market? Insurance companies? Automotive executives? Technologists? Should consumers be allowed to customize the moral dashboard of their cars so that their vehicles execute moral decisions that are in line with their own preferences?

1. https://motherboard.vice.com/en_us/article/j5a8d3/self-drivi...

This is just FUD, and it's got nothing to do with the parent comment, which was talking about the lives that will be saved by widespread autonomous vehicles. Not sure why you decided to respond to that comment with your own variation of the trolley problem but you're being downvoted because the comment wasn't relevant.

The fact that developers of autonomous systems have to consider the ethical and moral implications of their work is well understood, but these systems will still be much safer than human drivers.

"Not sure why you decided to respond to that comment with your own variation of the trolley problem but you're being downvoted because the comment wasn't relevant."

Because it's the only aspect of autonomous cars that worries me and parent was talking about how much safer it will be. I know it's not rational from a statistical perspective and actually safer than human drivers, however on an intuitive human level it's hard to stomach that I may be algorithmically placed in harm's way.

That’s the only aspect that worries you?

I’m far more worried about the infinitely long tail of edge cases, overhype of AI and artifically inflated trust that comes from it, increased car usage on the road when people would rather be ferried than take a ferry/public transit, decrease in ability to manually control machinery as a population... these all seem like certain critical issues that will affect masses rather than hypothetical trolly problems that will seldom occur but are easy to obsess over.

Provided that it's statistically safer, I'd rather be algorithmically placed in harm's way than placed at greater risk by drunk or distracted drivers.

>on an intuitive human level it's hard to stomach that I may be algorithmically placed in harm's way.

For the opposite, see "I, Robot".

Spoiler warning.

I initially thought of the main character's backstory (survivor's guilt over an algorithm rescuing him because he had a higher chance of survival than a little girl), but the main plotline is also about people being "algorithmically placed out of harm's way".

No self driving car thread is complete without someone spouting trolley-problem stuff.

Until AIs are significantly more intelligent and capable than humans (and hence capable of solving this problem better than any of us could now), they'll do exactly what humans do in a crisis: Say OH SHIT and stomp on the brakes while trying to steer away from anything solid.

If stomping on the brakes isn't the appropriate reaction, then the car was going too fast in the first place.

Clearly you live in a world without corners, rain or semi-trailers behind you.

I live in a world where a car striking a pedestrian at any speed under 20mph has a 99%+ chance of not being fatal.

I'm curious what you think about what a human is going to do in that situation. My guess is that they'll more or less pick randomly from the options you presented, plus a few more options like:

1. Do nothing at all because they're too distracted by their cell phone. 2. Do something worse than nothing due to panic and a lack of practice in such situations.

Personally I'm excited that these options will not exist for self-driving cars.

I'm also excited that we can talk about how the machines _should_ choose between the other options -- with human drivers we don't get that choice.

It’s not a terribly difficult dilemma: the car will save me, because I won’t buy a car which wouldn’t do that, and neither would anyone else, so that’s what we will be able to get.

The only way to change this is by legally imposing a certain decision tree - and then it’s also considered “solved”.

Apologies if these long cut and pastes are not in the spirit of HN, but I've saved this comment I found that succinctly explains why the AI trolly problem as it were just isn't something we should concern ourselves with at our current technological capacity. Double apologies for the comment's biting tone, but I found it entertaining.

Source: https://reddit.com/comments/9rav8y/comment/e8fps5m?context=3

What the hell environment do these non-technical types think the cars are going to operate in. Are they driving through day-cares or something?

These endless moral arguments are coming from non-technical non-business types who are desperate to try to contribute to the revolution that is selfdriving cars.

The very few technical types who make these moral pronouncements are identifying themselves as having a skillset so far out of date that moralizing is all they have left to "contribute".

It boils down to some simple "moral" choices. Cars are on roads, people are not. Some people will accidentally end up on roads. The car will have the wisdom to try to predict people being idiots and do its best to avoid them.

But very much like right now. Idiots who jump in front of cars are going to be darwin'd if the car has no easy option to avoid them. If the a child jumps out in front of a car, it too should, shall and will be darwin'd if the other option is injury to someone who wasn't a nitwit. I have exactly zero interest in being in a car that would say, "Oh, the occupant of my car is older than the nitwit child that just jumped out into traffic. I am now going to drive off a cliff to save the moron."

I don't care if a crate of orphans spills on the road. I hope my car will swerve if possible. But if any option involves risk to me, then a crate of speedbumps is how it shall be.

Why such "moral" choices? Because they won't be moral choices. The technical limits of weighing these things is not going to happen any time soon. The car will stay out of situations where it is at fault such as driving on sidewalks; after that it will do its best to mitigate for idiots, and then its primary purpose will be to keep the occupants safe. Otherwise, you will have the car doing things like avoiding a blowing garbage bag or somesuch that it identifies as a 4 year old child and then driving into a tree to prevent the travesty of scattered garbage.

Arguing that trolley problems are irrelevant because the software can't identify whether the obstacle is a pedestrian or a garbage bag reliably enough to take evasive action which might increase risk to the vehicle's occupants is actually a pretty strong argument against self driving cars....

(I mean, I sympathise a bit with the Reddit user's contempt for surveys which ask whether self driving cars should avoid over "a criminal" in preference to other types of people as if this were the sort of thing humans were capable of doing, but at the same time if I believed autonomous vehicle programmes were being run by people with his "occupant first" mindset - and I'm sure they aren't - they should be shut down immediately and permanently)

The trolley problem is completely, entirely, 100% irrelevant.

If the contrived set of circumstances that would lead to such a scenario actually happen to you, and somehow the AI is advanced enough to analyze the various possibilities while not being advanced enough to avoid potential dangers in the first place, congratulations you're absurdly unlucky. Potentially unfortunate for you, but fortunate for the millions that won't die as a result of taking dangerous human drivers off the road.

The trolley problem as it applies to self driving cars was from one engineer's offhand silly remark that the media loved. The same thing applies to automated airport train systems. It's not something that will actually happen often enough for anybody to care; and in the few cases it applies, it will be handled by the law, just as with any other system that causes a death.

Good to know, thanks for shedding light on the topic. Airport automation/Airplane autopilot is a great example and you've changed my mind completely so I'm 100% for self driving cars.

No downvote from me, but this is apples and oranges. The parent is talking about probabilities and you're talking about possibilities emanating from a single possibility of unknown probability. Those are not directly comparable.

If the probability of the flying propane tanks is low enough, then I don't need to worry about the car's moral priorities.


>ings have gotten much safer the past couple decades so that number is somewhere around 0.75% or lower now.

On a statistical level, sure. On an individual level you can improve your person chance for survival greatly by not driving drunk and putting down your phone, the shitposting can wait. Also if you've made it past your teens your odds improve greatly.

>I want to say being injured in a car accident is somewhere around 20-30% lifetime chance.

That sounds absurdly high. 25% is 1 in 4. Assuming odds were equal or higher in the past then the average person has one grandparent that's been injured in a car crash.

Whatever the rates for injury/death are they're probably not evenly distributed among the population.

Actually, OP's figures are pretty much backed up by the figures you get when googling this question.

Chances to die of a car crash/ any kind of vehicle crash is about 1 in 100 according to here [0].

I can't find something similar straight away for injuries, but car accidents are ubiquitous and 2m people are injured every year (vs 30k fatalities) [1; cached]. So something above >20% seems plausible.

[0]: https://www.iii.org/fact-statistic/facts-statistics-mortalit... [1]: https://webcache.googleusercontent.com/search?q=cache:dnqo-l...

Some UK figures here:


mortality due to transport accidents / total deaths = ~2000 / ~500000 ~= 1/250 or 0.4%

> On an individual level you can improve your person chance for survival greatly by not driving drunk and putting down your phone

Let me just add sleep-deprived driving to the mix

Wow so 4.6 million people seriously injured in car accidents in the US in 2016. Given an 80 year average lifespan you are statistically at 110% risk of serious lifetime injury using the most basic statistics.

4.6M / 330M = yearly risk 1.39% x80 years = 111.5% chance of serious injury

I'm sure that includes things like whiplash though.

Probably more correct to do one minus the chance of no injuries.

So the chance of no injury is (1 - 0.0139) ^ 80 = 32.6%, making the chance of serious injury 67.4% over 80 years.

That’s not how probability works - the probability of getting heads when you flip a coin twice is not 2x50 = 100%.

The rates really are that high.

Funny that you mention grandparents: if you graph the crash rate by age, you get a deep U-shape with the very youngest and oldest having a much higher rate.

Doesn't pretty much any crash at a decent speed (even if the airbag doesn't go off) cause an injury like a bruise from the seatbelt or temporary sore muscles?

Yeah, but so does closing a door on your hand. I would hope that whatever we're considering an "injury" is something severe enough to usually require medical treatment (kind of like when someone say "mass shooting" you don't usually think of a double murder suicide). Bruising except in the most extreme cases does not require treatment. Similarly for whiplash, if it's less severe than having a stiff neck in the morning like many people get from sleeping on a crappy air mattress or couch bed I wouldn't count it I think that including such minor injuries is misleading.

Whatever we count as an "injury" should probably be more severe than the little mishaps we have with our bodies from going through life.

>Whatever we count as an "injury" should probably be more severe than the little mishaps we have with our bodies from going through life.

The fact that this is apparently an unacceptable opinion here is mind boggling to me

If we're basing out stats on unimportant injuries (minor bruises and stiff necks that go away in a day) then we're measuring something that we don't care about, injuries that do not meaningfully affect people. What we should care about is injuries that have appreciable recovery time (serious bruising, having a sore neck for multiple days, etc, etc) because those negatively affect people enough to be worth caring about. If we're including injuries we don't really care about in our numbers then something is wrong with our numbers.

I dare someone to actually tell me why you think we should not distinguish injuries that are so minor as to not appreciably affect people from those that do seriously affect people. Tracking minor and significant injuries in the same category misleads anyone who has to reason about or make decisions based on those statistics.

What about not driving late when bars are open or closing and people get behind the wheel tired and buzzed?

This is true but there are also less people on the road.. That being said there are drunk drivers at anytime on the road.

Wondering if those numbers contain a bit of an ambulance chaser skew in there?

I doubt it. Among people I know, I've known 3 people injured as pedestrians, one person who was killed as a pedestrian, and a couple of months ago I was injured as a pedestrian.

Cars are ridiculously dangerous. Something needs to be done about it. Good thing there's Waymo and other companies like them.

Holy crap, where do you live, just outside Thunderdome?

One injury & the friend who passed away were in Sydney. Another friend was injured crossing the road in Bangkok. I was hit by a taxi whilst crossing the road (legally, with a walk light) in Singapore.

In my experience, humans on average are just quite bad at paying attention. The only reason autonomous vehicles didn't surpass our driving abilities years ago is because we tend to make our roads unpredictable (road work, human behaviours, etc.) and it turns out that humans are able to quickly comprehend those situations.

So, oddly enough you do sort of live outside the Thunderdome presuming you're in Sydney.

Perhaps, but as of yet there is no real evidence that self-driving cars will decrease the total number of deaths. In the short term, the latest advanced driver assistance systems which do things like automatic braking are proven to work. We should be putting a greater focus on deploying those more widely by making them mandatory on new passenger vehicles, just like air bags and stability control.

Research into autonomous vehicles is great and might eventually pay off, but let's not get distracted from real incremental improvements that are possible today.

This will be a huge milestone for the Waymo team. I had not thought they would get here until 2025 at least, so I guess I'm overly cynical these days :-).

I hope that a successful launch in Phoenix will expand out to other locales because this capability is a huge win for those groups that would otherwise not be able to make these trips.

It is also remarkable that it is possible to replace a regular vehicle operated by a human as a livery service, with an incredibly complex machine and still make a business case out of it. That says a lot about how far computers have come in the last couple of decades.

Personally, I think that all the self driving efforts would do well to be tightly coupled to city planning going forward and we should be specifically regulating toward "driverless zones" where a grid system can handle all traffic, a city switch, effectively.

There are efforts to fit driverless cars to horrifically planned streets, but I don't see any cities approaching it from the other direction.

This is all said though because I freaking hate cars and the entire ecosystem surrounding them.

>but I don't see any cities approaching it from the other direction.

Can you blame them? There's barely money to maintain existing infrastructure, let alone rework it. It doesn't matter if it would pay for itself in 25 years, if that were the case we would be powered by 100% nuclear power years ago, investors don't bite on those numbers and republicans don't pass those budgets.

Maybe the driverless car players should work with the cities to improve infrastructure. They have all the capital, after all. However that logical reasoning isn't how the valley plays ball. Look at bird, look at uber, look at any 'pioneer' service entering these cities. They stomp their way into markets without trying to reach out to local governments and work with them, the city officials of course are annoyed and pass unproductive regulations, then the tech companies throw a tantrum and plead their pitiful case to their users. It's childish behavior that wastes everyone's time and money.

We are in agreement.

This is what should really be happening with the cities viying for the business of Amazon, Twitter, whomever.

I'm constantly flabbergasted by how short sighted both cities and companies are about the long term impact agreements will make on a given area. They think in terms of placating concerns via taxes but don't look at the true opportunity loss from not addressing future quality of life in an area. FFS tech companies should deeply be judged on this factor when being assessed for a new HQ, as they clearly are not currently going through a rigorous process which will play out in a decade plus.

Or for a fraction of the cost you could have driverless trains and move 20 ~ 50x more people. Oh wait, that already exists. It's existed for decades:


Self driving cars are totally the wrong direction we should be going in, and it's going to cost the US more in the long run.

I'm very pro-train and anti-car, but in many situations you still need a motorized means of transport for the last mile. I think on-demand self-driving cars could be a good match to solve the last mile (together with bicycles).

> Self driving cars are totally the wrong direction we should be going in, and it's going to cost the US more in the long run

This goes against the research I’ve seen, which suggests a self-driving fleet can more-efficiently transport a dispersed population like America’s than rail. (For the relevant distances.)

Also, anecdote: after Uber, I use the subway and regional rail system more. It solves the last-mike and “should I rent a car” problems.

The car drives you to the bus, the bus drives you to the train, simple. The train is scheduled to depart at the best time for all the buses to arrive at. Less waiting, less buses and trains that are half empty.

For this to work the self driving car needs to be expensive enough for people to care to organize their trips this way. So we will need to pass road tax in cities to account for the externality of congesting the roads. As these cars are driven by corporates with constant GPS monitoring this will be more easily enforced than ever

Pricing in externalities is job 0.

I want to go from A to B.

Do you want option a) be alone, £100, b) share car, £60, c) most efficient, switch to bus and train, £30

Yes, ideally driverless cars would evolve together with the cities they drive in.

Unfortunately, it would take a decade for the government bureaucracies involved to even get started with that effort, so we'll have to make do with just the vehicle side for quite a while.

Volvo was looking into that. They want to drive magnetized nails into pavement on some Swedish highways to provide a lane reference. That can be sensed through snow, and snowplows can be equipped to sense it, so it has uses beyond self driving.

"Launch" invokes a mental image of a rocket taking off, but that isn't really an apt metaphor. Waymo has already had vehicles on some routes move passengers without safety drivers onboard. They've been charging some EZ Riders for their rides. When they do 'launch', the service still will not be available to the general public. The major thing they'll be doing is unveiling their new brand, whatever it is. They've been gradually expanding the scale, range, and capabilities of their fleet in the Phoenix metro area and they will continue to do so.

If the self driving tech could drive the airport rental car, amusement park parking, and Las Vegas Strip shuttles, as well as other similar shuttling venues, then that would expose lots of people to the tech in fairly constrained environments. This seeds a lot of citizens to react favorably when these livery services show up later in their cities.

I’d wait for the service to actually exist, open to the general public, before congratulating them. If they were confident, they would announce a date or just announce that it’s open for business. This sort of “soon, seriously!” announcement is all about satisfying investors, regulators, and business partners who presumably are holding their feet to the fire on their earlier promises.

Which investors? They have Google at their backs, why do they need to impress investors? My reading of this roll-out is that they're first to some kind of service, and they've been first to many other milestones. However, due to wanting to prevent an Uber situation at all costs, they are taking it slow and careful.

I feel like a lot of people had overly optimistic ideas about self-driving cars and then these pessimistic views developed as a backslash.

That happens a lot - a technology is hyped for so long that people become disenchanted, then boom suddenly it arrives and it is amazing. Think flat screen TVs, touch screen input, powered flight.

I think a similar thing will happen with VR/AR when it eliminates the efficiency gap of remote work.

Yes. But the optimistic ones are right.

If you believe in yourself, you will succeed

And no amount of not believing in someone will cause them to fail

So the pessimists are just wrong. And sometimes it’s fun to be wrong in a fun way... no shade on pessimists. But they are objectively wrong.

I will be really interested to see what limitations the service launches with. From what I've seen, they have been extraordinarily hazy about what the cars can actually do (e.g., time of day, weather, locations, dealing with non-road complexity). That makes sense, of course, as it's valuable information for competitors, but I'm still eager to find out.

I'll also be interested to see how much this is them wanting to run a full production service versus an advertisement for their technology. Will it be more like Google Search? Or more like Android, where they maintain a small market share as a demonstration of where they want people to go?

I've been impressed watching them test in the Sunnyvale/Mountain View area (I see them on the road nearly every day, often multiple times a day). Among the situations I've seen them handle: bicyclist swerves into the lane in front of them; pedestrian starts crossing the road right as they're about to make a right turn; pedestrian jaywalks in front as they're stopped at a stop sign; boxed in and prevented from making a lane shift on a 3-lane boulevard; avoiding construction on El Camino; stuck behind a bus; U-turn in front of a car edging out to make a right; unable to merge while coming off the highway; unable to merge onto the highway; tailgated and cutoff while on the highway.

It's possible that they were under manual control at the time, but I doubt it. They have a distinctive way of driving while under computer control, which I guess I'd describe as "like a grandma" but basically involves acceleration, braking, and turns that are a bit smoother than any reasonable driver would make them and a top speed that is a bit more law-abiding.

It is California, so it's always sunny (well, smoky right now) and most of these were daylight, but I've encountered them in a lot of different situations and consider them safer than human drivers.

Haha, opposite of my experience. They have failed to navigate the left turn off rengstorf El Camino for the last 2 years (they always stop in the middle of the intersection).

A pedestrian on the corner next to a stop sign causes them to pause in the middle of the intersection for an uncomfortable amount of time, to the point where you aren't sure if you should just go around them or what.

On my bicycle and motorcycle, I don't trust them at all. I've had them come within inches of me on a turn, without slowing down a bit. Maybe they saw me and executed a highly efficient path avoidance mechanism, no way to tell, all I know is they operate around me in a way that fires all my 'watch out, this motherfucker doesn't see you' alarms while I'm on two wheels.

Ok, I'm being mean - it's a miracle they can go at all, let alone not hit people. Today at the aforementioned intersection, at the turn they always fail at, a bicyclist got ran over by a pick-up truck. The waymo SUV slowly pulled up to the intersection while people ran back and forth dragging the bicycle and his groceries and other debris away, and didn't run anyone over. Never stopped slowly creeping towards people, which was weird as hell, but hey. Handled a super odd situation like a champ.

>pedestrian jaywalks in front as they're stopped at a stop sign;

>A pedestrian on the corner next to a stop sign causes them to pause

I am struggling to reconcile how the pedestrian does not have the right of way in both these cases.

What action you take depends on what state you're in, and where the pedestrian is.

> Pedestrian crossing: 50 state summary

> The majority of states, however, only require motorists to yield to, rather than stop for, pedestrians crossing at uncontrolled crosswalks. Nineteen states require a motorist to yield when a pedestrian is upon any portion of the roadway. Louisiana mandates motorist yielding when a pedestrian is upon the same half of the roadway. Nebraska requires yielding when a pedestrian is upon the same half of the roadway or within one lane of the motorist. Massachusetts mandates yielding when a pedestrian is upon the same half of the roadway or within 10 feet of the motorist; and 20 states mandate motorists yield when a pedestrian is upon the same half of the roadway or approaching closely enough from the opposite side of the roadway to constitute a danger. In addition, in at least five states and the District of Columbia, bicyclists have the same or similar rights as pedestrians.


In my example (standing at corner), they certainly have right of way, whenever they plan on walking. Sometimes they're just hanging out, which is usually obvious to a human, but in a subtle way I can't describe, let alone guess at how to program.

In my example (jaywalker) they stopped and waited for the pedestrian to cross the street.

The pause times do seem longer than a human would take (I noticed this with the right-turn with pedestrian case as well), but I'd rather that self-driving cars err on the side of safe rather than sorry.

Agreed, endgame (all cars self driving) it's gonna be better. But when there's a mix on the road, I imagine a lot of awkwardness. I feel like they should throw a little LCD panel on each side of the car so it can just display a message like "yielding to pedestrian" or whatever.

Mercedes has spent some thought on this. They propose having LEDs in the front of the car that colour where it sees pedestrians, so you know if you've been seen.

Imagine the lawsuits when someone relies on what they saw on the panel and an accident ensues.

This is why we can't have nice things.

1. "Jaywalker" ("crazy walker") is a slur invented by the auto industry to remove the civil rights of humans not in cars.

2. Your example doesn't even seem to meet the slur's definition.

I am not sure why you got downvoted for stating a fact. About the automotive industry's aggressive lobbying to gain control of the streets in the 1920s, see here for instance: https://www.bbc.com/news/magazine-26073797

> I am not sure why you got downvoted for stating a fact.

Personally, I downvoted it for the "civil rights" phrase. It is unnecessary, hyperbolic, and inflammatory; these discussions tend to attract such comments and they derail rational discussion.

1. 100 years ago yes, but at this point the term is in common use.

What's considered jaywalking (crossing a street where there is no crosswalk) in USA is usually legal in other countries. In Netherlands I can cross the street almost anywhere I want, except if I'm too close to a real crossing.

See https://en.wikipedia.org/wiki/Jaywalking#Safety_consideratio... for the various unique instances.

In some cities in the USA, Jay walking is also waking across a crosswalk when you don't have the walk signal. Is this a ticketable offense in the Netherlands?

FYI the definition of jaywalker is "a reckless pedestrian who crosses a street illegally". This sounds legal.

As a New Yorker currently living in California, this is something I've had to grapple with.

In New York, right of way is pretty unilaterally given to road traffic. Outside of otherwise controlled interactions (signaled crosswalks), pedestrians are expected to give way to vehicles. And vehicles (to borrow maritime terminology) have a stand-on duty--they are expected to maintain speed and direction.

So it doesn't matter what posture pedestrians adopt, even when they are standing a full lane into the road (people commonly queue to cross the street in the parking lane, rather than on the sidewalk, leaving the sidewalk clear for people traveling along the street). Cars are expected to proceed---and you can bet they will be reminded of that fact by those behind them if they slow unnecessarily. Furthermore, New York's robust jaywalking culture means that these people are all peering upstream, watching for a break in traffic.

In contrast, California has very much opposite expectations. Cars are expected to yield to pedestrians in pretty much all situations (and jaywalking is rare). It turns out if you exhibit really any of the above behaviors in California, drivers tend to interpret that as telegraphing an immediate intent and request to cross the street.

When I moved to California, I was initially perplexed and frustrated by how it seemed like cars would randomly stop anytime I happened to be standing by a crosswalk, even when I didn't have any intention to cross, or when I was happy to let them pass by first. Lots of awkward "no, no, please proceed" gestures ensued.

Eventually I realized the big cue I was giving them was that I was habitually watching traffic. I now take care to be visibly not looking upstream when standing at a crosswalk. That seems to have solved the problem.

So I submit to you that the difference you're picking up on between pedestrians who do and do not have an intent to cross is whether they are paying attention to traffic.

Your example perfectly illustrates why I prefer the California method. On my morning walk or bicycle ride to the train station, I have to go through easily 30 lighted intersections. Having to analyze traffic for a gap 60 times a day sounds exhausting. I much prefer just zoning out at the red light, listening to a podcast or whatever, then doing one quick left/right check when my signal comes on.

I don't have to analyze anything, I don't have to negotiate with a driver, don't have to guess speed, I just go when it's my time to go.

>I am struggling to reconcile how the pedestrian does not have the right of way in both these cases.

A pedestrian jaywalking does certainly not have "the right of way". That doesn't mean the car doesn't still have an obligation to avoid hitting them. The rules of the road are redundant for a reason, one party not doing what they're supposed to should not cause a crash.

If the car is at a stop sign no person walking in front of them could possibly be jaywalking.

The Rengstorff-El Camino intersection seems incorrect. If you are turning left from Rengstorff to El Camino it's not a left turn arrow, but actually when it's green, the cars on the parking lot on the opposite side have a red light. I've been confused by this because it seems like if there's a car wanting to go straight on the opposite side and no turn arrow, you shouldn't turn left, but that's how that intersection works.

That is indeed wrong, wonder if there's some public feedback mechanism for reporting non protocol conforming intersections :)

The City of Mountain View has a public works issue tracker system, which you can use to submit a request to the city engineers: https://clients.comcate.com/newrequest.php?id=128#

Keep in mind that both El Camino Real and freeway ramps are maintained by CalTrans, not the city. But the one time I submitted a issue about a freeway off-ramp they forwarded the report to CalTrans for me, so it's a good place to start.

Also that intersection is literally the border of Los Altos and mountain view, and is only "partially" managed by Caltrand. I dunno what that means, it's what the lady on the phone told me.

That intersection has always been horrible. Our office looks straight onto it so it's our favorite thing to just watch and wait for the inevitable (usually tame) accident.

"...you aren't sure if you should just go around them or what"

It just now occurs to me that autonomous cars could have status lights. So that an observer can better predict what's gonna happen next.

Kinda like the purpose brake lights and blinkers serve. Flashing yellow lights indicates the car is confused, thinks there's a safety risk, or whatever.

An expanded visual vocabulary.

hazards during hazards are fine but I don't think adding more status light with proprietary meaning would really help.

on the other hand a led matrix with pictograms would be great

> pedestrian on the corner next to a stop sign causes them to pause in the middle of the intersection for an uncomfortable amount of time

This is a really interesting point to me, and I think it's one of the major weak points for self-driving cars: they have to pass a kind of Turing test.

You have a good example of something we do all the time when driving: try to read the mind of the driver. Don Norman wrote a book titled, "Turn Signals are the Facial Expressions of Automobiles", but it goes well beyond that. We use the "body language" of driving to infer driver intent.

I think the best short-term case is that people make excuses for the dumb robot as you do here. But a very plausible outcome is that we get an "uncanny valley" effect for robot cars where people resent and are creeped out by something pretending to be human but falling short.

What they need is to tell the people around them that they've been seen, maybe an led strip that glows in your direction, green for "congrats, you're very predictable", blue for "what the hell are you doing", and black for off "haven't seen you yet"

I watched a driver-intervention near-crash today on 16th street with cruise where the car sped up and swerved into oncoming traffic across a double yellow to avoid a car backing up that was obvious to the human driver. It's only luck that the cruise car didn't end up in a head on collision.

Do you mean "smoother than any unreasonable (human) driver"?

> From what I've seen, they have been extraordinarily hazy about what the cars can actually do (e.g., time of day, weather, locations, dealing with non-road complexity).

Not sure this is 100% related, but the driver-assistance technologies in my new 2019 vehicle such as the lane keep assist, blind spot sensors, adaptive cruise control all suffer varying levels in inclement weather. My lane keep assist is completely unreliable in the rain, the collision detection sometimes misfires (but thankfully doesn't brake me) or mistakenly thinks an oncoming car around a bend is in front of me, etc. There are thousands of edge cases I can think of, just with my rudimentary understanding of the technology (namely lidar) are thinking to my self there is just no possible way they have come to a point where they are safe for general use.

Although inclement weather is certainly a challenge, comparing the sensors in your car to the sensors on a commercial self-driving taxi is of limited use. The sensors on your car are designed to be affordable first and foremost, and likely cost hundreds (or maybe in the low thousands on very high end systems) of dollars each so they can sell the vehicle. The self-driving taxis are designed to work no matter what, cost be damned, so they can build out the technology and let the cost come down later with scale. There are single sensors that cost tens of thousands of dollars on some of the A-list players in this space, and sometimes several of them. The sensors cost more than the car, not to mention the cost of the computers themselves potentially costing more than the car (this varies by company).

As a rough rule of thumb, the companies testing level 4 autonomous vehicles on the road today are paying around $100k-200k per vehicle in hardware costs alone.

You're comparing apples to oranges. The hardware and software running on fully self-driving cars is much more sophisticated than mere assist aids on regular cars. Your car does not have LIDAR, for starters.

I was making two separate points about current assistance technology and the capabilities of LIDAR. (I think the technology in Tesla's are pretty good - Elon tends to think LIDAR isn't necessary, who is to say he is wrong and their opinion is any more or less valid?)

In either case, it isn't hard to imagine, the fundamental algorithms are the same -- recognize lane markings, adjust steering wheel, that kind of thinking. What do you do in the case that the lane markings have been worn down on the road? Freshly paved roads that haven't been painted? I could go on ad nauseam but point remains: too many corner cases, lidar or not.

I'm not sure why you are getting downvoted, tbh. LIDAR helps, but is certainly not sufficient for many of the reasons that you mention.

There is a reason that these will be severely geographically restricted for a significant period of time.

Teslas also like to accelerate into road barriers. It is hard to consider them a legitimate competitor in the self driving space right now.

Yes. Tesla pulled their self-driving claim last month.[1] Their self-driving video, from 2016, was apparently the one trial that worked out of a large number of tries.

[1] https://www.theverge.com/2018/10/20/18000884/tesla-full-self...

> Their self-driving video, from 2016, was apparently the one trial that worked out of a large number of tries.

Are you speculating, or is there a source for this point?

I don't doubt it, just wondering, as I think evidence for this would be fairly incriminating.

> combined with eyewitness reports of multiple days worth of filming, it’s clear that at a minimum Tesla’s system had many attempts to practice the route before recording a truly autonomous run.

Interesting. I'm a little surprised this did not get more coverage.

> too many corner cases

No more than a human needs to handle.

Humans are allowed more failures than auto autos.

My ACC in a 2018 Golf does not seem to work well in a steep ramp uphill with stopped cars. Or perhaps I did not have enough balls to wait a little bit more to see if it would stop after it first accelerated as I approached the stopped cars.

Current driver assistance systems don't use lidar. It's still too expensive. Some of them rely on stored maps that were created using lidar on dedicated scanning vehicles.

They've actually been pretty transparent about the system's limitations. Check out Waymo's application to conduct driverless testing in California: https://drive.google.com/file/d/1oZ9zwmpfzZvuGVX7MeUGvUSW6Vq... Specifically the section on the vehicles' intended operational design domain.

The cars will operate 24/7, in a small, well-defined section of metro California. They can handle light rain, but won't work in heavy rain, snow, flooded roads, or one-way mountain roadways.

Though that document also says that they won't be charging passengers, so I guess it's a little out of date.

Interesting! Thanks. That's from Feb 2017, I see. I'm not sure I'd call "burying key information in one required regulatory filing" pretty transparent, but I suppose people differ on this.

Different parts of that PDF are from different dates. Not sure which part you're seeing from 2017, but the Application for Manufacturer's Testing Permit is dated October 11, 2018.

I suspect your perception of how transparent they're being might differ depending on where you're located; Waymo's disclosures so far have been mostly in the form of communication directly with the communities they're testing in; not as blog posts or press releases to national media. That document, for example, contains copies of emails that Waymo sent directly to city officials in the places where they planned to start testing.

Thanks. The 2017 date comes from the last page, which first I took as a "received" stamp, but it's only a stamp for the last document.

Yes, I am definitely basing my understanding of their transparency with the public on what they have said to the public.

So the truth is Google is expanding its pool of test users, removing their NDAs, and starting in the easiest place in the world to operate a driverless car.

It’s certainly progress, but I would have appreciated a less click-bait headline.

How is that a click-bait headline? Waymo is launching a commercial driverless car service, just as the headline says.

Of course they're starting in one of the easier places to operate driverless cars. Why would they introduce more risk than necessary to the launch of an extremely complex new technology?

The title refers to this very specific bit:

> It will operate under a new brand and compete directly with Uber and Lyft.

It wasn't clear until now what the plan was, if they would start their own service or run their cars inside the Lyft/Uber network. The fact that they will start a new service with a new name is pretty big news, and therefore not click bait.

> Waymo is launching a commercial driverless car service, just as the headline says.

When this launches, will the cars have drivers in them?

That’s how you can tell if it’s a driverless car.

The idea that driving in a well-tested Phoenix suburb is even the same activity as Manhattan, as some seem to think, is comical.

You start in the south where the weather is easy and work your way north. I think the hope is that by the time they get to Manhattan, global warming will have improved the weather.

It isn't just weather. Phoenix is probably the closest thing we have to a perfect city for this tech. In addition to the clear and precipitation free weather, its geographic location gives it the most daylight hours of any major city in the country. The roads are almost all big, straight, flat, relatively new, and easy to navigate. There is not much of a pedestrian culture. Drivers are not super aggressive like they are in some other highly congested cities. There is almost no native wildlife that would pose any threat to a car. And I am sure there are other things I am missing.

>There is not much of a pedestrian culture.

Arizona has the highest rate of pedestrian deaths in the US. Pedestrians have apparently been trained to keep off the streets.

It's also not immediately clear if this will be in the city of Phoenix itself. The earlier testing was in the suburb of Chandler.

The weather is the least of your worries in a place like Manhattan.

Pretty large chunks of the country are a lot like phoenix suburbs.

Indeed although many have much more inclement weather than Phoenix. To be clear, this is good progress. But expansion may be slower than many think.

And also lots that are an incremental step away. Gotta start somewhere.

A non significant chunk of trips are in places that don't look like them. If I still need a driven car for 10 or 20 percent of trips, I'm not going to ditch in favor of driverless cars.

It's a service, like Uber, you're not expected to ditch the car at this stage

If it gets cheap enough I at least could take the self driving car to the nearest train stop and go downtown via train. She got grocery shopping in the suburbs. However, shopping be better only a single stop. Maybe we could solve that problem though by having some kind of bin that gets automatically stored at a grocery store while I'm shopping, loaded with my groceries and then put into the next self driving car...?

Of course it's not the same, but it's also not incredibly easy. I drove around Phoenix for about 7 hours recently. Definitely not the easiest roads, nor the most considerate drivers. I'm curious to see how these cars will figure out all the weird lane closures, double-lane turn-right-on-red, blind intersections, etc. I almost got creamed turning left near the Heard museum, and there was a bunch of confusing construction downtown. And the highway merging can be... exciting. Not saying it's the most difficult driving in the states, but it's also not the easiest.

Manhattan also has a robust public transit system, making it feasible for a person without a car to get around. Driverless taxis are probably disproportionately likely to be used in suburbs and other places without that infrastructure.

Yet, that isn't where taxis are primarily used today. Why would driverless taxis be much different? Maybe they'll be a little cheaper (though Uber drivers are probably only making minimum wage--if that) but it's not clear why that would transform behavior.

> Yet, that isn't where taxis are primarily used today.

Huh? Taxis are for sure used way more in NYC than out in the suburbs. They benefit significantly from density. Out in the suburbs you can't even catch a taxi unless you call one to you.

The popular way of getting around in the suburbs is driving your own car. In cities mass transit and taxis are much more common, especially because parking is so expensive that it makes more financial sense to hail a taxi for the occasional trip.

That’s what I was trying to say. Dense urban areas are where taxis are used the most and those will be hardest for self driving.

Indeed. It's less urban areas that have poor public transport that need this the most. The bay area for example is extremely hard to get around without a car, but something like Seattle? Very different.

Environments like phoenix are also where cars are the most useful.

Lfyt and Uber must be having a heart attack right now.

This was always a clear threat. While Uber is also working in this space, it always seemed very clear to me that once you “solve” driverless cars, creating a network similar to Uber is relatively trivial, assuming human capital is not required 1:1 as it is with Uber.

In theory, to the customer, it doesn’t matter if a human or machine is driving, so you’re no longer dealing with a two-sided network, making adoption substantially easier, especially for a company like Google who can deploy massive capital.

The thing I never understood is how Uber’s investors rationalized this and thought it wouldn’t happen. Did they think self driving cars would not be a reality? Did they think Uber could get there first? Did they think there is still a profitable enough gap between current situation and the dreiverless future to hedge in case Uber doesn’t get there? Is the Uber brand and tech worth so much that they’ll get something back as an acquisition for whoever does get there first?

> Did they think Uber could get there first?

Seems like exactly what they thought. Remember Uber acquiring Otto and Levandowski? AFAIK Travis was pushing hard for autonomous driving but accidents like the fatality in AZ and missteps like Levandowski's fraught relationship with Waymo led to deemphasizing autonomous driving for the time being. Dara's main job seems to be to take the company public, which requires scaling and fixing the margins on the products they already have in place.

I don't work at Uber and perceptions are my own, so curious if any Uber employees agree.

At the current cost of a self-driving car (~$250K), it would be ridiculously expensive to build a fully self-driving ride-sharing fleet [1]. So expensive that I doubt it's possible to make real money that way right now.

Self-driving cars are a long game until companies can bring down the costs. It might be years before that happens. Uber has the advantage of being able to ramp up self-driving cars as part of their existing fleet until then.

Edit: Math mistake

[1] https://qz.com/924212/what-it-really-costs-to-turn-a-car-int... [2] https://www.uber.com/newsroom/company-info/

I believe that the cost to produce a self-driving car will be coming down, but let’s look at the $250k price tag.

Over 6 years, that comes to $114/day. 1 ride per hour at $5 per ride would hit $120/day.

Assuming a 2 mile ride, that would be 105k miles which is well within the car’s usable life. At 3 miles, it would be 160k miles which is still within a car’s lifespan. It looks like Uber is $1.35/mile with a $2.10 base fare and $1.85 fee. A 3 mile ride should be able to get $5.

Plus, it’s really about the long run. Operating margin might be negative for a bit, but the cost of the technology and manufacturing will come down significantly. Still, even today, I think $250k just isn’t that bad. I think most drivers will do a lot more than $5/hour in gross revenue.

It would be really expensive to make a fleet, but the economics are so compelling, even at high prices. I mean, Americans are often spending $35,000 on a car. If a self-driving vehicle can service the needs of 7 people, it can be cheaper than car ownership.

While the price of driverless tech might be high now, the prices of car ownership and human labor are both very high as well and only one of those three prices is likely to decrease over time.

Isn't 6 years quite long for a high tech car? I wonder what maintenance would cost.

Factor in the cost of gas and you're in the negative already.

I suppose most of these vehicles will be electric, and cost of charging is lower than cost of refuelling...

Your operating cost appears to be 100% capital costs. What about, you know, operating costs?

This is trivial.

Driverless cars are "always on" so it's not an apt comparison because you're going to always get more from less. Besides, ridesharing is already subsidized right now, so bottom line isn't an issue in the market.

Likewise, ride-sharing with self driving cars is even easier since if you own a self driving car it can work on it's own while you are not using it, which is most of the time. There's no reason a similar asset sharing model won't spring up after economies of scale.

You could also argue that it's cheaper to pay workers with their own textile equipment, rather than spend the CapEx on a weaving loom, yet here we are.

It's not trivial when we don't even know what sort of regulation will be required to maintain these vehicles' road legal status, let alone banking on future economies of scale. Would it be a simple license fee? Perhaps an inspection? An age requirement on the vehicle? Or perhaps the worst (and very possible) outcome, mandating that an undistracted driver be sitting behind the wheel at all times, ready to take control.

It would take one trivial piece of legislation to turn this technology into a souped up cruise control rather than the world changing technology that its backers insist that it is. How could you even fight that sort of regulation without your argument boiling down to 'we promise we don't need human oversight.' One lazy 'think of the children!' retort later, and it's banned faster than mango juul pods.

I don't understand the investor confidence here. To me this seems like basic research, critical for future technological developments, but a cash sink with no guarantee for profitability. Is this just a rat race between the giants throwing cash at this?

The likelihood of such a legislation on a national level is unlikely. Instead, you will have cities that refuse to adopt self-driving, maybe because they were burned with a tragedy (with no regard to statistics), maybe to protect human jobs.

Large companies can adopt in other cities, or if need be, other countries eager to change their quality of lives.

Precisely. Regulatory adoption won't be much different than how government has responded to ridesharing today. Highly fragmented but typically open.

If this is trivial to you, what do you consider moderately difficult? What's sufficiently difficult that you have to spend a single day thinking about it? How about an open problem requiring careful planning and months of effort?

It's like the word "trivial" has no meaning anymore. All I can tell from people using it these days is that they're very confident in what they're about to say.

What I consider non-trivial is tangental to comparing the cost/benefits between autonomous driving and ride share services. The non-trivial part of this is [nearly] solved, as is evident by this Waymo soft launch.

The self-driving cars may be available to give rides 24/7, but if 90% of the demand is 0800-0900 and 1600-0200, they may end up sitting idle for twelve hours each day anyway.

You probably need over 100x the vehicles at 5pm as you do at 4am. You have to find a balance of how much CapEx is worthwhile -- probably more than you need to cover the 4am shift, but far less than needed for the 5pm shift.

Factor in the cost of a driver, and the $250K price tag will look super cheap if not sensible.

Right now, rideshare companies keep ~40% of the ride revenue (not including vehicle payments). Which means, 60% goes to the driver. Now if you consider operating costs (fuel/gas/insurance/tolls/oil change/maintenance) at 30% (of the total revenue), there's still the remaining 30% that the driver takes home. Even if you think the driver is working 80 hrs a week, that's not even 50% of the total number of hours an autonomous car can work (24x7 = 168 hrs a week).

With a fully autonomous car, that remaining 30% and lower maintenance costs combined with lower insurance costs will mean probably close to 40% of the current ride revenue will be saved. If you operate the car for 160 hrs a week (8 hrs for fuel/maintenance) then it'll probably be ~60% of ride revenue (at current rideshare rates) that'll be pure profit.

As cost to produce these cars (basically it's the separate technology package added to regular cars) goes down, they can continue to keep the service super affordable while recovering the upfront investment in the $250K car in a few years. Instead of having 10 Chrysler vans on car dealer lots/storage, have one $250K car in service a day after it is manufactured. After that maintenance costs may increase slightly, but the rest is pure profit, which creates a positive feedback loop to

reduce vehicle price --> reduce final fare paid by passenger --> attract more users --> more vehicles needed --> Leads to larger scale production and reduction in price--> repeat.

At some point it'll be cheaper than public transit and then owning and driving a car becomes a need only in regions / places / conditions where autonomous cars wont work. Until then, non-autonomous cars ensure good margin for autonomous cars to thrive.

FWIW, I believe Uber has zero advantage with it's existing fleet. All they have is a large number of indentured drivers (due to subprime loans given to them for their cars). In fact it is a disadvantage for Uber, because as fares for ride share goes down, it starts making lesser sense to operate those cars. they'll just stop making payments and return / trash the vehicle.

I don’t think you need same number of self driving cars as drivers and the ratio might be way higher than 1:3. Plus I think you have a math error and the total cost in your example is $250B which is doable for Google.

>So expensive that I doubt it's possible to make real money that way right now.

I don't think anyone is interested in "mak[ing] real money that way right now", they want to get a foothold in, or capture a swath of, a market which definitely will make them money in the future. A tremendous amount if they play it right, even if it means operating at a loss for the next few years.

People on this site have as recently as a few months ago predicted that full self-driving was more than fifteen years away.

Deploying a preliminary taxi service in one city in a place with no weather is not even close to “full self-driving”.

When they’re driving around with no safety drivers in Manhattan in the winter, we can talk about solved problems.

The goalposts have successfully been moved.

I'd rather have a machine driving. I've had some frankly dangerous taxi or rideshare drivers.

Until you run into a large static obstacle...

I believe you're referring to an issue that Teslas have experienced? Waymo cars are way more sophisticated, both in software and hardware. They use LIDAR to see large static obstacles and don't ignore them.

We now are going to get Uber complaining that Google is taking their jobs, and not being regulated like they are, and how technology shouldn't put people in the streets.

Humans drivers could be necessary to... just lend the car. Not for their driving skills but for the money they invest. It’s much easier to let car owners take the risk of investing in a car, maintaining it, repairing it (for the moment they don’t repair themselves), cleaning up the Saturday night mishaps. It could be hundreds of billion dollars to populate the world with Google-owned self driving cars.

Alphabet is making $8 billion a quarter in profit currently. They'd probably love to have some worthwhile investment to sink that money into.

> Humans drivers could be necessary to... just lend the car.

Unless the humans are losing money by doing so (which may be th case, but probably doesn't make a sustainable business model), the car service is ultimately fully paying for the car for the period it uses it, so it doesn't lose anything by doing so up front. It increases the up front capital requirements, but Google has more cash on hand than the market cap of some Fortune 500 firms, so that's not really an enormous issue for Google.

> It’s much easier to let car owners take the risk of investing in a car, maintaining it, repairing it (for the moment they don’t repair themselves), cleaning up the Saturday night mishaps.

Well, except humans lie about the last bit to get extra reimbursement, whichnpisses off customers. That kind of thing and other driver frauds is a compelling reason to get independent driver-owners out of the loop.

Consider the fact that car manufacturers don't generally operate taxi services. For the same reason that Toyota sells Priuses (Prii?) to taxi companies, Waymo may choose to sell its technology to car manufacturers and via them to taxi services.

Given all the problems that Uber, in particular, has had with their humans, I'd expect that customers would prefer a non-human driver.

Your margin is my opportunity - amazon

Your non-automated workforce + margin is my opportunity - google

Amazon owns Kiva which automates warehouse management

>> then they lfyt you, then you win

Seriously though, I think mobility network effects can be overcome because they're inherently local, and they're "attackable" because there's decreasing returns to scale. Average distance to nearest ride tails off pretty sharply. Obviously not as weak a moat as scooters though, because it's still very capital intensive.

The first-mover also incurs some regulatory cost and rider suspicion the followers might not, too. Still, it's an enviable position to be in.

I sincerely doubt it. This "driverless" program will still have a back-up driver and thousands of dollars of additional equipment on board. Along with, I am sure, several limitations on which specific locations this can work. Uber operates at a global scale and in plenty of cities (think central London or south Mumbai) where a driverless car share program is decades away (if viable at all).

The premise of Uber's (in particular) valuation is that ride-sharing is a network effect market, and leaders can maintain leadership.

I think as that's a scary premise to be hanging the company on. First, it's too big a market for one company to control. This isn't Facebook. Second, self driving can make this into a one-2-many market without needing Uber. Third, rides have prices. This isn't Google. A cashed up competitor can price war their way into market share. Uber won't likely be able to leverage their position in the way digital monopolies have. Prices mean price wars.

If I were Uber, I would be looking for an immediate IPO.

It definitely could be a problem for Uber’s IPO story. I’m sure Waymo’s progress was foreseen.

If driving becomes a commodity, the bigger long term problem is the mobile phone being the gateway to calling a ride. The users may just use Google/Apple maps or the voice assistant.

Uber launched a self driving car service in Phoenix like 2 years ago. It recently hit and killed a woman and the governor of the state shut it down.

So... These will have to have people in them until they are 100% effective which many people think is still 5~ years away. Until then Waymo just have very very very expensive (to them) ride sharing, which is really unlikely to scale.

They already don't have human drivers in them.


I think it will help the blind, and people with seizures, so I hope it gets figured out. I do some driving for people with visual impairment, and I enjoy helping, but this would be best.

TLDR; Waymo will be spinning up new company next month exclusively for app based rides (name is being kept secret). Fares will be competitive with Uber/Lyft. They did year long test drive for 400 customers in Phoenix. These same customers would be offered initial access. Other competitors still remains far behind in terms of disengagement metric.

This will mark the first ever commercial self-driving cab service.

There have been rumors of Waymo preparing 60,000 self-driving cars. So next year (or two) could be game changer.

That is not a rumor. Google and Fiat announced the deal. Remember, the entire game for self-driving-car journalists is finding new and creative ways to ignore the fact that Waymo is obviously far ahead of everyone else. They ordered 62000 vans, but Cruise's big advantage is partnership with GM. Uber kills people while Waymo has been shuttling hundreds of people around Arizona for months, but keep writing about how it's a horse race.


Why spin up a new company? Is it to limit liability?

Waymo does the R&D for the autonomous driver.

Then separate companies build a business around each use case to monetize the driver.

1) Consumer rideshare

2) Courier / delivery

3) Corporate / fleet / trucking

4) License driver tech to OEMs

5) Lease self-driving vehicles outright

6) ???

It looks like Waymo might become provider of tech to sub-companies that operates specialized segments like self-driving cab, self-driving food deliveries, self-driving trucks, etc.

Chandler. A wealthier suburb of Phoenix.

While I was there, the local TV channels ran segments a couple times a day. Very well done.

I assume it was paid editorial content.

Applications are open for YC Summer 2019

Guidelines | FAQ | Support | API | Security | Lists | Bookmarklet | Legal | Apply to YC | Contact