Apple scientists disclose self-driving car research | Hacker News Reader

Apple scientists disclose self-driving car research

reuters.com

204 pointsrbanffy173 comments

There was a time when I thought of regulations as progress killers, but I find myself shifting on that and getting nervous every time I hear about self-driving tech.

It's great stuff, but I really do not want my family to be the brave eggs we had to crack to get to the omelet of a driverless future.

I now _like_ that someone somewhere has a book listing chemicals that aren't allowed to be shipped on an airplane, and I wish there were a way to jump right to that mired, bureaucratic point in car tech.

Similarly, I used to view lawsuits as a menace to free enterprise, but I'm a tiny bit comforted by the fact that everyone working on car tech knows they're extremely exposed to litigation if they aren't diligent.

The idea of code from Apple or Google driving on my roads is switching me from classical liberalism to "help me, big government!"

The reason why consumer/enterprise software is so lightly regulated is that very few people die when mistakes are made. (Embedded, especially medical, is a different story: https://en.wikipedia.org/wiki/Therac-25)

Meanwhile, cars (and roadways designed for high speeds without proper pedestrian accommodations) have been the #1 cause of injury deaths for kids for generations: https://www.cdc.gov/injury/wisqars/LeadingCauses.html

I would love to see the numbers you pulled out of thin air for your statement on deaths caused by software. I've met at least one developer who quit the profession because they were writing code that "would probably lead to someone's death."

> they were writing code that "would probably lead to someone's death"

Versus cars, where I'd be shocked if there's a single mass-produced model that hasn't killed someone.

As of 2015, there were 9 car models without recorded deaths - https://www.nbcnews.com/business/autos/record-9-models-have-...

I'm not sure how statistically significant the numbers are, but it's something!

More complete driver death rate statistics can be found here.

http://www.iihs.org/iihs/topics/driver-death-rates

If I'm reading this correctly, those cars nave no "driver deaths" that means, that no one has died while driving those cars. This is a bit like a shotgun, which has potentially killed many people, but has not killed its owner.

Imagine if everyone was only allowed to drive those 9 car models.

Then there would be deaths. The stats are skewed significantly towards vehicles of which there are more of so you'd always have to discount this figure by how many vehicles there were of that type to begin with and to be even more accurate how many passenger miles were driven with those vehicles.

I'm a pedestrian. I don't drive. I'm sure the cars don't have air bags on their bumpers.

Actually some Volvos do have pedestrian air bags!

(But your point still stands, of course.)

> The reason why consumer/enterprise software is so lightly regulated is that very few people die when mistakes are made

So few people die because computers are nearly unregulated and mass manufactured. Computers are cheap and safe because we allow the market to function.

Incorrect. Look at applied computing for medicine/healthcare, those things are strictly regulated and analyzed because of their potential to harm. A 0.001% overflow in your twitter app makes you restart your phone, a 0.001% overflow in your pacemaker means you die. Immediately.

I realize these regulations exist, my point is they come at a cost and that at some point the costs can become prohibitive. As I mentioned below the argument isn't just theoretical, it's why the term "medical tourism" exists.

I believe reasonable tradeoffs could be made in theory, but I also believe the appeal of regulatory capture is too alluring for wall street to pass up. And when the people who need these devices can't afford it - well, that's when operation Too Big To Fail goes into effect.

the argument isn't just theoretical, it's why the term "medical tourism" exists.

Can you explain that in more detail? What's the connection to regulations? Does medical tourism exclusively happen from countries with heavily regulated medicine to countries with light regulations?

I would have assumed it was more about the distinction between state-run same-care-for-all healthcare systems versus commercial pay-to-play systems. If you have the money and you need the care, of course you're going to pay for better service if possible. But that seems orthogonal to regulations.

Both arguments are valid at the same time. People do absolutely shop around to skip waiting lines in state-run systems or save money by getting private treatment cheaper (UK citizens getting cheaper IVF in Spain is a common one I'm familiar with), but it's also sometimes true that certain treatments or medications that have been made available in one State that might not yet be certified for clinical use in another. New stem cell based therapies (the merits of which I have no idea...) in some South American nations is one example of the latter.

That's true. I still don't see the connection to self-driving cars, though, and the idea that regulations "come at a cost and that at some point the costs can become prohibitive".

GP is concerned about "regulatory capture", but the UK/Spain example you gave is about states being faster or slower to allow access to certain medical techniques, not corporations exploiting the system to lock in their market dominance.

The point about regulatory capture is secondary to the increase in cost. But there are patients going from high cost to low cost countries (linked below). There's also "big pharma" that is using IP regulations to limit what choices consumers have. I realize IP is another can of worms but it's still an example of regulation that is having the exact opposite of the desired outcome.

I would totally be for "sane" regulation, where autonomous driving components can be inexpensively tested and certified. I fear the more likely outcome will involve keeping costs prohibitively high. Considering how many people die on the road each year it would make sense for the state to sponsor a non-proprietary solution. Off the top of your head, in terms of the number of lives that can be saved per dollar - can you think of a better investment?

My feelings about autonomous cars is largely influenced by how terrible humans are at driving. At this point I'd rather trust a decent self-driving car than the average human.

*https://www.npr.org/templates/story/story.php?storyId=162941...

There is regulation based medical tourism. For example, the Czech Republic is the only country in the "westernish" world which allows a special kind of IVF in which the nucleus is taken out of a near menopausal woman's egg, and placed into the mitochondrial fluid of a young woman's egg, and then fertilized using sperm from the husband, making a baby with genes from 3 people. Women from more western countries pay good money for this treatment.

Whether it is a good thing that it is legal here is questionable. Doctors seem very skeptical of this 3 genetic parent baby approach, and I don't know if it is because it is inherently risky, or if they are just conservative regarding the ethicalness.

I think the UK allows this as well.

Based on what? What deaths is the lack of regulation preventing? How would they be more dangerous if they were regulated?

There's lots of unintended, expected, negative outcomes from regulation.

The counter culture is unregulated experimentation like https://www.ncbi.nlm.nih.gov/pmc/articles/PMC3298919/ An impetus to circumvent these regulations, in the extreme case to save lives (or just to be deviant). Regulation is like a wet blanket that can mold, but is effective at putting out a fire in a pinch, if it's big enough. Regulatory bodies never seem to have a path-to-dissolution process (the core of my problem with regulation) and are largely implemented in such a manner that they grow, self-perpetuate, morph and fester corruption for countless generations (e.g. FDA vs smoking tobacco).

Admittedly, I'm biased by my beliefs. Despite being heavily regulated, I think much of the cancer epidemic is caused by auto and oil/fuel, with some contribution by Monsanto Roundup-like products. The independent studies, in these heavily regulated industries, are consistently squelched, influenced, or made unavailable (trade secrets!).

Regulation is not a silver bullet.

> How would they be more dangerous if they were regulated?

Not necessarily more dangerous, but more expensive. Medical tourism anyone?

One study by RAND says: "Autonomous vehicles would have to be driven hundreds of millions of miles and, under some scenarios, hundreds of billions of miles to create enough data to clearly demonstrate their safety... Under even the most-aggressive test driving assumptions, it would take existing fleets of autonomous vehicles tens and even hundreds of years to log sufficient miles to adequately assess the safety of the vehicles when compared to human-driven vehicles" https://www.rand.org/news/press/2016/04/12.html

That being said, RAND released a study this month, targeted at policymakers, that models how many human lives would be saved given the introduction of non-perfect autonomous vehicles, including a little javascript widget to try out various settings: https://www.rand.org/blog/articles/2017/11/why-waiting-for-p...

Crucially, this study models scenarios such as AVs being _half_ as safe as human drivers, yet still resulting in saving more lives over time. I think the question then becomes, who will be responsible for the missing 50% of safety when somebody gets killed by one?

Edit: Another question: if somebody you loved was killed by a self-driving car, would you be comforted in knowing that the technology was in aggregate saving the lives of other people? I have no answers to these questions... it's a hard problem.

> including a little javascript widget

This is great. You can make any model say anything you want by tuning the parameters, so we needs researchers to stop forcing their preferred parameters, and let consumers check the robustness of the conclusions against ranges of parameters

> if somebody you loved was killed by a self-driving car, would you be comforted in knowing that the technology was in aggregate saving the lives of other people?

and if somebody you loved was killed by a human-driven car, would you be comforted in knowing that the technology was in aggregate was being held back for further testing?

> if somebody you loved was killed by a self-driving car, would you be comforted in knowing that the technology was in aggregate saving the lives of other people?

with a handful of centralized systems responsible for every accident everyone will be permanently panicked and outraged at its purportedly uncaring, greedy overlords.

> if somebody you loved was killed by a human-driven car, would you be comforted in knowing that the technology was in aggregate was being held back for further testing?

counterfactuals need to be near and specific for people to give them much thought, see:

http://www2.psych.ubc.ca/~schaller/Psyc590Readings/Roese1997...

http://www3.canyons.edu/faculty/rafterm/0%200LLI%20LOVEandLO...

> Crucially, this study models scenarios such as AVs being _half_ as safe as human drivers

See, this is the main reason why AVs are great: accidents are caused by speeders, drunk, tired, inexperienced drivers, morons driving while high on pot or ecstasy, that is, those who are MUCH worse than an average driver used as a yardstick to measure the performance of the autonomous cars.

Replace them by an imperfect machine lacking common sense, and the overall driving skill still goes up.

The "replacement" implies that people you mentioned want to be replaced. I don't think that's the case; speeders will just disable AI and crank up the speed themselves. People who are reasonable enough to activate AI when they are drunk will probably just order a taxi now, and so on. The only way to have that "replacement" is to force the autonomy on drivers, but that's not gonna fly well on a market.

> The "replacement" implies that people you mentioned want to be replaced

Not the entire world is America. Oftentimes "want" is replaced by "forced".

> Another question: if somebody you loved was killed by a self-driving car, would you be comforted in knowing that the technology was in aggregate saving the lives of other people? I have no answers to these questions... it's a hard problem.

We already have analogous situations with things like airbags. Airbags unarguably save lives, and yet they've also unarguably killed kids.

How many of those deceased children would have survived the collision if no airbags were installed?

Self-driving cars preform predictably under a set of conditions. The primary problem arises when the car, for whatever reason, doesn't correctly detect that it's in an unfamiliar condition and continues moving as normal. Humans are intelligent enough to notice when something is "off".

For a machine as heavy, large, fast, and close as a car, limited intelligence is not enough.

Another quesion: if somebody you loved was killed by a driver's human error (quite possibly their own), would you be angry about the delays in making autonomous cars available?

I understand the reluctance to trust new technology. If we are talking about the risk/cost of new developments, we should not forget the baseline costs we pay today.

Self driving car tech will be the only method that can provide a hundred fold decrease in driving fatalities. Slowing down progress has a cost too.

You could also just reduce driving deaths by having better public transportation and reducing driving overall.

Sounds like it would even be better in the long run.

Self driving public transportation could provide the benefits of both.

Driverless trains have been here for a while. They're great. With platform-edge doors, nobody can fall or jump onto the track, not to mention that they increase brake cooling efficiency and decrease air resistance (less air to push).

But of course you weren't talking about trains.

I see this sentiment a lot on HN that self driving cars are going to fix everyone's transportation woes, but I'm not convinced that they will. For example, they're never going to be particularly efficient at really high speeds, due to tyre wear (due to heat) and air resistance (which can be reduced by driving in a convoy, but not enough). Rail (or maybe hyperloop, who knows) will still be much more efficient for high-speed travel, not to mention much faster. So maybe the cars will all be going 40% faster on highways than nowadays, which is nice, but still only half of what rail can do, and even less for hyperloops.

I don't buy the "it's gonna fix everyone's commute", either, because at the same time it's supposed to fix parking. So now you have loads of cars driving out of the city to park at the same time as you've got just as many driving into it. Sure, vehicle-to-vehicle communication and superhuman reaction times will allow for increased density on roads, but is it enough to offset the increase in traffic? And is the increase in miles driven and resulting road maintenance cost factored into the cost calculations?

Damn right I want cars to drive themselves, and it's going to be fantastic. But I don't think autonomous cars are the panacea that many seem to see in them.

I agree with your scepticism about solving the commute traffic with a self-driving silver bullet. The research for traffic flow [1] says that, with human driven cars, a slice of a 4m wide lane will pass about 1000 people/hr in cars, about 7000 people on foot or bicycle, and about 10,000 people by bus.

That means that to get competitive flow with buses, self-driving cars need to deliver a tenfold improvement in flow. Since in traffic cars are quite tightly packed already, you probably can't get much from putting them closer together, so even if self-driving magic gives you two doublings of flow on its own you still need to more than double the speed limit to compete. I can imagine this working only on roads dedicated to self-driving vehicles, with no pedestrian crossings, so now self-driving cars benefits only materialise if they eat up a chunk of the public space.

You might get a bit of improvement from vehicle sharing as well, but it seems like a really complicated way to attack a problem that already has an OK low-tech answer, at least in high density areas.

Where I live the census says that about half the commute journeys in the area are less than a few miles, for which cars of any sort seem ridiculous to me, for the able bodied. For those who aren't able bodied or have a long journey all the able bodied car users on short trips are using up a scarce resource that they need!

Throw in the health benefits of active transport, and the health cost of particulate emissions (still a problem with electric cars, since a decent chunk comes off tyres and brakes) and it looks like a wash to me. Then again I am one of those awful bike people, so maybe I miss the point.

[1] Litman 2017, Brun & Vuchic 1995 etc.

Outside of NYC, in the U.S. public transit in my experience takes two or three times as long to get you where you're going. Until the typical experience gets better or people get poorer, the resource efficiency won't matter.

I guess part of the issue here is that cars are a kind of coordination problem - if nobody else is using a car on the roads, cars are quite a good choice. They are fast and easy, and go where you want.

However, each person who uses a car creates some external costs (a high traffic factor / unit of person-flow) which are borne by all the other road users. Once people choosing the "defect" strategy (cars) have enough numbers, the "cooperate" strategy (not cars) gets broken for other solutions that use the road network, so the system fails to a bad equilibrium if there are enough defectors in the population.

Congestion pricing would solve that, and go nicely with robocars too.

(I'm not a fan of cars; I don't drive one myself. But I don't expect people to accept any less convenient and flexible solution.)

The common American is getting poorer, but just like junk food and lack of exercise, people still drive. Other modes of transit may be much safer and healthier, but unless the common person is encouraged (whether that be with incentives to not drive, or tolls for driving) quite a few people will still drive.

There are a number of US cities that get most workers into the urban core besides NYC despite poor to middling infrastructure at best, but driving infrastructure dwarfs everything else, bike lanes are added only to diet roads usually (resulting in shitty lanes), and most areas see 60% to 70% of their land tied up in (mostly empty) parking, which is a disaster for every mode of transport.

Not sure what rail services are like where you live, but in the UK it's about 50/50 whether it is faster to travel between major cities by car or by train. Trains can go faster but this is often counterbalanced by needing to slow down to come to a stop, etc. There will always be places that can't be reached with public transport and even in cities like Paris renowned for their public transport it is often necessary to change several times as part of a journey.

Many light rail services e.g. BART and the London Underground are completely crazy at rush hour and there are many obstacles to increasing frequency. A big challenge of self driving trains is ensuring they are able to stop in time if an obstacle is present, and this relies on fairly similar technology to that of driverless cars. Self driving cars would reduce traffic because they would reduce accidents and allow cars to sort themselves into slots automatically rather than waiting at traffic lights.

You have a good point about rush hour traffic being increased by cars driving outside the city to park, although cars could park more efficiently in city centres if multi-storey car parks didn't need to allocate space for humans.

I'm confident in saying there's zero chance that self-driving cars, on their own, could replace the London Underground. Zero.

The Underground (and commuter rail that feeds into it) carries a vast number of people every day. There's just not enough road space to get that many cars in and out, even if they were 2x or 4x more efficient.

If you want to fix the Underground (although it works fairly well so it's not clear why drastic changes are needed) you need to start thinking about buses and bicycles, staggering working times to avoid rush hours, moving jobs outside the city so people don't have to commute in at all, that sort of thing. None of those actually needs AI.

Self-driving cars (and especially self-driving buses and minibuses) can certainly be part of the solution, but they're not the solution.

Where a train can move 20k+ people an hour in each direction, a standard road moves around 1k. Self driving cars will only make sense as chauffeur replacements for Uber, Lyft and Taxis, as even with a tripling in capacity, you'd still be wholly uncompetitive with rail. Nevermind the inefficiencies of enlarging roadways :P

> If you want to fix the Underground (although it works fairly well so it's not clear why drastic changes are needed)

The population of London continues to increase. Central, Northern, Jubilee lines are often difficult to get on during rush hours, and even harder to stay on. And that's in the winter, summer time's a whole other ball game...

Self driving cars are certainly not the solution, but that's not to say there's not plenty that needs fixing on the London Underground.

https://www.youtube.com/watch?v=FOF_XIylzS8 still rings true, albeit with less BK these days than before.

I don’t think I’ve ever heard someone claim driverless cars will fix everything

Mass transit specifically trains have serious problems I don’t think there are solutions for. The first is that they make centralized points of failure. Broken trains can bring a whole city to a halt. The second is they are the perfect germ delivery system.

> just

It's not as easy as you make it sound. Americans still want cars.

Most Americans need cars.

America is a vast land compared to European countries. Public transportation can only go so far. Some American suburbs stretch more than 80 miles out of major urban centers.

Most Americans need cars because American urban planning has been for many decades built around cars.

This, of course, is a solvable problem.

(It has nothing to be with how big the US is, it has to do with how US communities are deliberately engineered.)

Solvable by what? Demolishing the suburbs? Building new cities in the middle of nowhere? A time machine?

Some improvement can come from just improved planning in the cities we already have. Relaxed zoning for density, exclusive corridors for mass transit, etc.

Definitely. Many suburbs need to return to farmland, we have paved over most of our prime farmland for shitty, rent seeking suburbs.

Dense cities are cheaper to live in, cost less to maintain for everyone involved, and give people back time that is currently stolen from them during the average commute. You also end up with better infrastructure, lower taxes, and usually a healthier and happier population in dense cities.

Suburban and exurban slum development with 1 to 3 stories and a 1 to 2 hour commute is destined to die, its just a matter of how and when. Detroit is one failure mode, while Livingston, TX is another (tax base could never support the infrastructure built), though the latter case is very common.

Everyday people commute from all over England to London, via train.

...because the marketing department tells them they do?

It's certainly a factor, but I think claiming that is the only factor is an oversimplification.

Or you could do both. The right tool for the job.

In the US alone, 35000 people die and more than 2 million are injured every year in car crashes. Technology is our best hope to reduce these numbers. Airbags, blind spot detection, lane keeping, emergency braking can all contribute. Self-driving level 5 will not suddenly appear and work everywhere in every situation. It’s a gradual process. Another thing to keep in mind is that insurers will not insure these cars for cheap until they are safer than humans.

Technology has helped a great deal. I'm saddened that enforcement hasn't played a bigger part, though. What I see around me on a daily basis is pure complacency. People switching lanes without turn signals, drivers with half their car in the bike-lane, tailgating, people who have their blinker on for several minutes (how did they forget? are they blasting music in their car such that they can't hear the clicking noise? are they that distracted??) etc.

It feels like my fellow drivers simply don't care about the human life that surrounds them. They'd rather text their friends than pay attention to the several-thousand-pound machine they are operating. In my opinion, obtaining and keeping a driver's license is just far too easy in the U.S., but I guess there is no hope to fix things on this end for two reasons: (1) losing your privilege to drive a car is a "death sentence"... how will you get to work and pay the bills?, and (2) police departments are facing budget cuts, so enforcement probably will not increase.

We have hope that self-driving cars will remove these negligent/distracted drivers from the road, but at the end of the day, what does that say about people that they can't take operating a heavy machine responsibly? Do we have any future tech that would emerge where ordinary people can threaten each other's lives?

...

Also, road design in the U.S. appears to care more about throughput than safety. Traffic circles are far safer, but they aren't used enough. Anecdotally, near my hometown there is a 6-lane road intersecting an 8-lane (!) road. That is the most dangerous intersection in the town, and IIRC the county! It's no wonder why having so many lanes of cars hurling themselves through an intersection is dangerous, so why are people allowed to design these roads in the first place?

> (2) police departments are facing budget cuts, so enforcement probably will not increase.

I would assume the opposite. Speeding tickets are a way for cash-strapped police departments to raise money (or justify their ROI to whomever controls the budget).

There are also other ways to successfully lower the death count, but as the recent NY Times column [0] says the American culture is working hard against any improvements based on restrictions.

[0] https://www.nytimes.com/2017/11/19/opinion/america-is-now-an...

Is it fair to compare the US to Europe?

US has an extremely different driving culture than Europe. ~13500miles/year for the US; ~8000miles/year for Europe. ~800 cars per 1000 people in the US; ~550-600 per 1000 in Europe.

Also, a larger number of people in the US simply couldn't live life without a car. I'm sitting in my hometown, a mere hour drive away from the largest city in the US, and I'd have to walk 45min to the nearest food. Work would be a 10 mile bike ride each way. This is New Jersey, the most densely populated state in the US.

The seatbelt thing is just us being dumb though.

Their metric is "deaths per billion vehicle miles travelled", so yes, I think it's a fair comparison.

It naturally scales up and down to match the amount of driving that happens in different countries.

I'm only deeply familiar with the US and UK, but the contrast there is pretty stark. I routinely see people on their phones or driving drunk in the US. Those do happen in the UK but they're a lot rarer.

The cars in Europe are generally much smaller too.

There’s a simple explanation:

You’re getting older, more conservative and more averse to change. It’s ok, it happens.

As part of this you’re also likely over-estimating the danger and underestimating the benefits.

I’m sure every self-driving car death will (and should be) scrutinized heavily. But remember there’s between 30,000 and 50,000 auto deaths every year. I’m betting self-driveing cars will do better.

"I’m betting" Thats the problem. How do you know? When you have family, and people you care about, thats not 'conservative'. Thats the reality of being human. If 'change' means that consumers become the testing grounds for unsafe technologies, then yeah, I'll be plenty 'adverse to change'. These technologies do not require people to die to be made.

Trying to find some good statistics but looking around it seems a majority of accidents end up being caused by human error, distracted driving and drunk driving. AI wouldn't have any of those so I'm personally confident that it'll have a much higher level of safety.

The trouble is when we swap "human error" for "machine error". Because humans screw up all the time but a machine is going to fail in a way that'll likely be very obvious to us. That is going to be take an emotional toll, I think, and will be a hurdle to get over for self driving cars.

No one wants to be told "sorry but your family died because of a bug in the car's software told it to make a left into a semi".

> Trying to find some good statistics but looking around it seems a majority of accidents end up being caused by human error, distracted driving and drunk driving.

This is a big one that most people seem to miss: A self-driving car doesn't have to be better than an alert, conscientious human driver to be better than 90% of the actual complacent, distracted, tired humans on the road.

"AI wouldn't have any of those so I'm personally confident that it'll have a much higher level of safety."

AI and perfect AI are different things.

Agreed but that doesn't make my statement incorrect :)

I said AI wouldn't have the issues of human error, distracted and drunk driving. It opens up all new types of errors, sure, but it wouldn't have the biggest killers of people. So unless we screw up pretty badly here it almost has to be lower mortality rates.

If we were to slow down the progress of self driving cars, would that not also be a bet on human drivers?

> "I’m betting" Thats the problem. How do you know?

I don't know why I have to point this out, but: betting is different than knowing.

Consumers have become the testing grounds for unsafe technologies since the invention of the Model T Ford.

But there is so much low-hanging fruit in automated driving! People are really pretty bad at responding to road hazards, anomalies (dog runs into road) etc. Also people are terrible at attending to repetitive tasks. And people fall asleep.

Its arguable (measurable) that automated drivers would be safer than human drivers. Resisting this is simple FUD?

> Resisting this is simple FUD?

There's some psychology at work as well. Say self-driving cars cut the road accident and fatality rate in half but in the remaining incidents, half of those are driver error but the rest are all due to hardware and software problems.

I think a lot of people would have a problem if half of all fatalities are due to system glitches even if the net result is fewer deaths and I don't think that's entirely unreasonable.

It will also be somewhat new ground. As I've noted before, there are relatively few examples of consumer-facing technologies/products where properly maintained and used (or sometimes even misused) products kill and maim--and people are just fine with that because "stuff happens" and the cost/benefit is reasonable.

Drug interactions are one of the relatively few areas today and even those often lead to lawsuits. (And the US even basically indemnifies drug companies against side effects from standard childhood vaccines.)

Maybe self-driving car companies should be indemnified for anything other than gross-negligence until the technology matures a little.

That will probably have to happen at some point.

My big concern is we’ve seen how reckless companies have been (Samsung with Note 7 batteries, Facebook/Twitter with the election, AirBNB(?) with reviews about unsafe places, Uber with basically everything...) I don’t want those companies making self driving cars without strong oversight. ‘Move fast and break things’ has done a lot of damage over the last few years.

I don’t want them to be allowed to apply that strategy to cars. We couldn’t trust their judgement for simpler things, the risk is way too high for cars.

Cigarettes are probably the best example, they look a lot more people than cars.

They're an example although I would argue that, in this case, "properly used" equates to not used at all. Also cigarettes were the subject of long drawn-out legal cases and a substantial settlement so they're something of a unique product.

> I think a lot of people would have a problem if half of all fatalities are due to system glitches even if the net result is fewer deaths and I don't think that's entirely unreasonable.

Why is it not unreasonable?

Because you’re preventing 30k deaths that would have happened but causing 8k that wouldn’t... it seems fair for people to be mad about that 8k.

Maybe we shouldn’t have rushed into things until that number was way smaller.

‘Diverless cars cause 30% of road fatalities’ is not a good headline.

It’s still a huge net reduction in deaths anyway you look at it, and there are huge gains in other areas (speed, covenience). I remain very unmoved by that argument.

I’d rather be hit by software than a human. At least I know we did as much as we could to alleviate the problem.

I’m not arguing it’s perfectly logical. If people behaved that way the sunk cost fallacy wouldn’t exist.

I think it would be a real problem. If the death count becomes non-trivial before the death count from people becomes relatively trivial I think there’s will be a big mess in the media.

It's all in the spin: ‘Drivered cars cause 70% of road fatalities’ is not a good headline for drivers.

That’s not new. Right now that number is 100%. People just accept it as status quo.

But people would be outraged by the 30% number.

> It's great stuff, but I really do not want my family to be the brave eggs we had to crack to get to the omelet of a driverless future.

The problem with this analogy is that it's based on the assumption that we are not currently cracking any eggs. We are. Hundreds of thousands every year. Introducing AVs right now wouldn't increase that number at worst; realistically it would likely decrease it. Your family of eggs is at a far greater risk because we are holding back. That's how atrocious humans are at driving (and making decision to get behind the wheel drunk) - we evolved to roam and climb trees at less than 10mph, so you can't blame us. Stop overestimating our ability, though.

Regulations may slow down innovation, but they also protect against companies who try to socialize the externalities of their business activities.

To me, full laissez-faire capitalism is a lot scarier than a system in which everything is overburdened with regulations.

> There was a time when I thought of regulations as progress killers, but I find myself shifting on that and getting nervous every time I hear about self-driving tech.

IMO, this notion applies to garage tinkerers and should not apply to billionaires or their corporations

The reasons should be obvious, but a garage tinkerer has a finite reach in whom his work can impact

The same isn't true of the wealthy elite and corporations. Who very plainly have the ability, in modern society, to impact just about every human on the planet.

It's as if, on one hand, we recognize government must be reigned in because of the immense reach of their power, but ignore the same problems when it's Bill Gates.

Propaganda is a powerful thing.

How so you feel about the 30k people who die each year the tech is delayed? Or do you really think the litigation system (where currently car manufactures almost are never liable) is properly tuned to minimize the time until that is achieved?

How do you feel about the 30k people who die each year the tech is delayed?

I'd say, let's ban using your phone while driving and drinking before driving like many other countries did 10 and 20 years ago. That could save 15k of those people right away.

Over the past decades, phone usage has surged while deaths have fallen or stayed flat (modulo the miniscule uptick that was all over the news). Distracted driving is dangerous, but phones aren't contributing nearly that much (and it's mostly illegal).

Drunk driving is super deadly, but it's already illegal. More importantly, self-driving cars will be vastly more effective at getting drunk drivers out from behind the wheel than all the legistlative in the world.

Driving while using a phone is illegal in most areas, but still very common, just like speeding -- it clearly isn't being addressed very seriously. And standards for "drunk driving" vary a lot; sometimes you're allowed two or even three drinks before driving.

Another commenter here linked to an excellent NYT article exploring this: https://www.nytimes.com/2017/11/19/opinion/america-is-now-an...

The US had one of the world's lowest rates of driving deaths in 1990, but since then other countries have pulled ahead significantly. Why is that?

I could imagine a million reasons. Maybe the US has a higher natural rate of accidents but deaths have fallen earlier because we implement sophisticated safety tech earlier. But once safety tech reaches diminishing returns, other countries pass us.

Maybe the US has a higher natural rate of accidents

Natural?? Roads and cars are all human-made! Are you saying the "deaths per road mile travelled" metric is wrong or biased somehow?

"Natural" in this context is just shorthand for "at fixed level of safety tech". You can substitute "baseline" if you want.

Why would the US have roughly the same baseline safety as everywhere else in 1990, but be 2x worse today?

The idea is that the US leads in safety tech adoption (which I believe is just an empirical fact) but that safety tech is reaching diminishing returns (conjecture), so that as time goes by it's tech lead helps less and less at lowering deaths per accident (relative to other countries), while it's accident rate remains higher.

This is really not that important. My point is that there are a million ways to explain the data besides the one you are suggesting.

Did you read the article? It picks out specific differences, like the rate of seat belt usage -- higher in other countries than in the US, even though self belts are mandatory just about everywhere.

"The US has laxer laws and doesn't enforce them as strictly" is a very simple explanation that fits the data well. It matches what I personally have seen in the US and UK, though of course that's just anecdotal.

I'm not sure if the US is ahead of the curve at all in safety tech -- things like ABS and airbags, right? The US is definitely behind the curve in adoption of traffic circles / roundabouts rather than 4-way stops.

You're correct that there could be some other reason why the US could have a higher "baseline" rate of traffic accidents; but what could that reason be? Any suggestions?

Possibly the US has a higher proportion of young drivers?

I’m pretty sure we have laws for that everywhere in the US

One way to roll out the tech is to do a hybrid system. Instead of selling self-driving tech as a convenience, the tech could be used to backup human driving. If the car sees I missed a stop sign, it could stop.

The goal shouldn't be a self driving car better than humans, but a self driving car better than a human driven car with automated safeguards.

This is the case with a lot of implementations, and it’s really lot a good idea if you want people to actually use it that way. Ignoring the effect that kind of system has had on pilots, in the automotive industry it’s just going to put poor systems into use under the guise of “well you shouldn’t be counting on it anyways”

Either something works well enough to be used throughout an entire use case with less failure than a human (driving on a stretch of highway, switching lanes, getting off at an exit, etc) or it isn’t

This all hinges on the unknowable of how safe the competing system will be. And how long it will take to roll out - a decade at absolute best before 50% of journeys are done with no human attention.

Launching unsafe self-driving cars will delay the rollout as soon as the first big accident happens.

The brave eggs will likely live in 3rd world countries with roads even worse than those in the US and drivers magnitude more insane.

Most likely, it will take a while until the locals understand the value of the training data to tax / sell it, so the vendors will have it nearly for free.

I'm still uncomfortable with lawsuits, but mostly because they're really inefficient ways to enforce regulation. In most places, legislatures add an area to the remit of a regulator; in the US they create private causes of action.

Hate to break it to you, but your family is much more likely to get killed by their own mistake or the mistake or someone else (e.g. a drunk driver) than a self driving car.

>I really do not want my family to be the brave eggs we had to crack to get to the omelet of a driverless future.

Simple way to avoid that:

Make a law that requires the CEO CTO and all the engineers and their family Who built the driverless cars To all drive around in their driverless cars / bike / walk around those driverless cars For a non negligible amount of time

No reason they should take the upside and public takes the downside.

I know you just want to keep your family safe, me too.

But consider my case. I have NO traffic tickets on my record, but I recently started taking a medication that could cause sudden blackouts. I am still allowed to drive.

I also have an ADDICTION to mobile games. I try to limit my playing to stop lights, but if traffic seems light, I will try to get away with crushing a few candies while driving. Studies have shown this is MORE dangerous than drunk driving, but I don't feel like it is because again, I don't even have any tickets on my record.

Yes, I have had a few close calls, and hit a few curbs, but I'm not stopping what I'm doing anytime soon. In fact I just got a mount for my dashboard so I can watch Youtube videos while I drive, so the problem is getting worse.

I consider myself a good person, and a responsible person in general, but I will probably be responsible for a horrible accident at some point before I actually change my behavior. I'm hoping the self driving cars get here before that happens, because my attention holding skills and driving ability are on a serious downward trajectory.

EDIT: To the people who got upset by this. I am sorry. This was Trollish behavior. I took patterns that I am seeing, and tried to personify them to make it more effective.

You can see from some of my past posts, I have real concern about where smart phones are taking us, and I took my argument too far. I really did lose someone close to me because they were hit by a driver that was looking at a dick pic and not paying attention to the road. This has permanently effected my perspective on this issue.

Also... the medication I was referring to was a normal SSRI Anti-depressant. These don't have any warnings about "operating heavy machinery", and are totally legal to take while driving. They don't cause black outs, but actual seizures in some people.

> I am still allowed to drive.

Are you certain about that? Most jurisdictions I know have laws that make it illegal to drive in when your ability to drive is impaired - even if it's cough medicine that could make you drowsy - let alone blackout.

edit: you are most likely breaking the law, that you haven't been caught yet isn't being "allowed to drive". Since we share the road with you, I'll repeat parent's appeal but directed to you: "help me, big government!"

It's likely that a blackout drug would have a provision in the paper saying "you are not allowed to drive while under this medication". It's quite common.

I appreciate the honesty, and apologies if I am missing some sarcasm, but please do the world a favor and at least throw out the mount. You are aware of how negligent that behavior is. If you kill someone, it is inexcusable.

GP comment is likely writing fictionally, as a hypothetical counter to the GGP.

A big difference is accountability. If you get into an accident and kill someone while playing a mobile game or taking a medication that says, "do not operate heavy machinery while taking this," it's clear who is at fault, and it is relatively easy for the state and injured parties seek sanction and redress through the justice system.

So what about an algorithm that has been given the regulatory green light and fails under some edge case or adversarial attack? How about something like the Tesla situation, where they advertise "self-driving" capabilities while tucking in some liability limiting fine print about how the driver should be alert and ready to take over control at all times?

Google says they want to be on the hook for their self-driving cars. (And they don't want humans to be able to take over in the first place.)

Sure. That way they can partner with insurers to offer lower rates to people who buy their tech. Pretty sound business strategy.

I understand that this is a pretty cynical view, and to be fair all the Googlers I've spoken with are true believers in their tech (perhaps a little too much). But don't underestimate the power of ego and financial incentives to constrain ethical/moral calculus.

Oh, it makes perfect business sense for Google to ask to be on the hook: less hassle for the customers means a bigger market for the product.

This is an inaccurate description of the Tesla system. When you use it, the car repeatedly tells you that you have to stay alert, and it forces you to have your hands on the wheel most of the time.

Then perhaps it should not be advertised as "autopilot".

Get your act together, do you even realize that you’re a criminal?

Read it again, they aren't being serious. Just illustrating different ways humans are dangerous drivers.

I think this is mostly a regional problem. In some parts of the world, you'd be arrested for using your phone while driving -- and quite right too!

This danger can be largely (unfortunately not totally) mitigated just by passing laws to prohibit dangerous driving, and enforcing them sensibly. It also helps if people agree that the rules are sensible and self-police them -- as happens in many countries with seat belts, for example.

I agree that self-driving cars should improve safety, but the idea that they're the only way to improve safety is just wrong.

If you'd like someone to chat with, my email is in my profile.

>yfw this comment is unearthed during discovery and you are f*ed in court

What's supposed to be the point of your post?

That police can't catch every criminal?

Stop it already. How much do you want to kill someone?

I don't see any particular reason to have more confidence in a government bureaucrat than in a lawsuit averse company, especially in the early stages.

The streets where I live have giant pot holes.

The government had the job of creating a flat surface. I got a surface with holes in it.

If we had a competent government they could probably just upgrade the roads nation wide with smart sensors that cars can use for guiding - along with a integration module for car manufacturers to use. But we don't have that kind of competency, so for now we're stuck with Google and Apple.

Every paper that comes out of Apple gets disproportionate coverage! If nothing, their secrecy is helping in generating good publicity.

I definitely agree with you, but have to add that my (amateur's) take is that this paper has actually contributed a lot. Not only have they shown marked improvements in detection accuracy, but also come up with a trainable model/framework for object detection/classification using only LIDAR point clouds (i.e. without the addition of additional sensor data such as RGB overlays). Not relying on a second (or 3rd, or nth) sensor or hand-crafted features is very impactful to the field.

I'm pretty sure that marketing is the purpose of their secrecy.

As this becomes more common that will cease to be the case unless it’s especially interesting.

I wonder if companies, like Apple, will segment their tech stack to extract areas where one could argue that it’s not ethical to have an intellectual advantage. Perhaps pedestrian detection is one such area.

If two autonomous car companies are competing, it makes sense to fight on many feature sets. It feels wrong to compete on “ability to detect a bike rider.”

Historically competition has been one of the best ways to foster innovation, and as a society achieving a good "ability to detect bikers" is probably something we want in an automatic car.

However, it is relatively wasteful since many resources need to be allocated to do the same thing several times. Alternatives could be : 1) deals between companies to not innovate in those specific areas, 2) a workgroup formed by companies working together on that subject 3) a government program to research that specific area.

> 2) a workgroup formed by companies working together on that subject 3) a government program to research that specific area.

Part of Japan's postwar economic miracle involved doing both of these things at once: "hey three giant Japanese companies making products using [some technology], here's a pile of money, if you each also contribute a pile of money, we'll run some R&D on [some technology] to make it better, more efficiently than if the four of us spent the same money separately, and you can all three use it to compete with foreign companies"

This is actually a field where competition is working as it should: cars are reviewed based on their safety rating which are common in the industry, car insurance is also based on that. The law is also very strong, there are strict road and driving regulation, so there is little chance for externality like "oh yeah, sometimes cyclist die around our car", ...

Really looks like the ideal environment to let the private sector figure it out by itself, no need for more government.

This is an interesting point. It's in every self-driving vehicle company's best interest to gain public trust in the industry as a whole. A high pedestrian injury/fatality rate from Google's self-driving tech reduces confidence in Apple's tech and the entire idea of self-driving vehicles. The incentives are there for these companies to work together to create highly reliable self-driving tech.

I don't think this is completely true. If someone actually has a meaningful advantage in safety, it's in their best interest to publicize that and use it as a wedge between them and the competition. Bonus points if you can use the ensuing backlash of your competitors killing people to write regulations that lock them out of the market.

Not necessarily what is best for people, but I don't think their incentives are as aligned as you make it seem.

Well known from the airline industry decades ago that you just don't talk about safety vs your competitors.

I would expect Volvo to do something like that, and repeat what they once did with the seat belt.

> repeat what they once did with the seat belt

Make it standard equipment a decade after competitors offer it as an option?

They were the first ones to offer a three points safety belt. That was the real breakthrough for consumer cars ;)

They also offered the patent for free.

I like the idea of this, from a practical point of view maybe the first step is sharing datasets ...

I agree with the sentiment, but the details will contain fractal devils. The most unusual pedestrians contain the most important learning experiences, but are also the most significant from a privacy perspective.

Edit: or, to put it another way, you don’t want to find out the hard way your machine vision system just mistook a drunk furry thalidomide victim for a wild coyote, but they might not be too thrilled to be on a database of “freaky humans”.

Google shared some of their oddities in the past: https://www.youtube.com/watch?v=Hsr3Fzi5clw (Including a lady in a motorized wheelchair chasing a duck with a broom?)

Legally, if you can see it from a car on a public road their is minimal expectations of privacy.

Depends on your legislation. In Germany you would violate the rights of third parties by filming the road with intentions of sharing it with other third parties while not also making the former third parties unrecognizable (black bar)

I doubt that applies to Moderate resolution LIDAR images which provide ~20 or fewer data points about someones face. https://cdn-images-1.medium.com/max/1600/1*vrB0Jrbt4Sjhkw2g_... Might allow you to detect a long beard or very large hat, but don't allow for identification without other evidence.

These questions will get very interesting when self-driving car manufacturers try to enter markets outside the US. It is highly desirable to have local training data, but any large enough effort to collect that data looks indistinguishable from mass surveillance. In Western Continental Europe this is bound to conflict with their expectation of privacy in public spaces. Meanwhile Russia and China might not be so thrilled about the USA having LIDAR scans of their entire country

They do compete on braking distance right now, isn't that the same?

Probably the only ethical solution is to develop everything open-source

If Apple cannot gain on their competition by being able to better detect bikers, what would be their incentive to put money into developing that technology?

Not getting sued for millions or billions of dollars? Not watching their reputation end up in the gutter due to negligence on their part?

That hasn’t stopped other companies in this space.

Having their technology not kill people?

So Project Titan is not really 'dead'?

I think Apple would much rather buy an existing car company than build one from scratch. I don't see them leaving that market because there is a lot of money to be made and they want to be there everywhere in your life.

Surely, if they killed it, they would still be in a position to release any findings.

Also, one could argue that given that this has been released, it probably lends more to the idea that it's dead.

I think project Titan was trying to build an actual car. Teams designing door handles and wing mirrors etc.

I think Apple are now focussed on trying to build the technology that would make a self driving car.

I think if they were successful in making the technology they would probably build their own car.

This is more like the effort on capacitive touch screens happening before the iPhone eg let’s develop a technology and see where it takes us.

It seems that they are following the same direction as Google's and focusing on the software and self-driving equipment as opposed to building a car from scratch.

See also discussion on the actual paper: https://news.ycombinator.com/item?id=15756020

Self-driving cars are high up on the morally-urgent technology list: car crashes cause more than 2% of annual deaths worldwide.

So good on Apple. I hope they get good PR on this, and I hope regulators factor in the very real human cost of gumming up the works. Don't get me wrong — they ought to do their due diligence, but they ought to do it swiftly.

how do you convince the midwest to use self driving tech? maybe start with nascar?

It's actually probably an easier sell outside of major population centers, especially if marketed as an advanced form of cruise control, which people already use on long, straight shots down country highways.

I think people would be much more inclined to trust self driving tech in these scenarios than in congested traffic with complex intersections and people unexpectedly darting this way and that with little to no warning.

That's more or less where the major auto manufacturers are all headed in the near- to mid-term. I realize a lot of people don't get excited by something that isn't on-demand door-to-door transportation. But it seems fairly obvious that full highway self-driving, especially limited access highway driving, is a much easier problem that is mostly solved today. It could end up being decades ahead of general-purpose everywhere/when autonomy. The only real question in my mind is when you transition from systems that are at least nominally assistive to hands off the wheel under certain conditions.

Detroit is a hotbed of self driving technology right now. The biggest automakers in the world have not been sitting on their hands.

As a guide to midwesteners, it might help to think of them as any other consumer who is driven by economics and convenience when it comes to car purchases.

How do you convince the valley to use self driving tech? Maybe start with putting beds in the back seats since they can't afford exorbitant housing prices.

I think you start with a a commercial product, which doesn't exist right now.

You have a car show up at their house every morning that's heated and ready to go? Even in car-centric environments people will want to save money and be able to use their phone instead of drive.

nascar != midwest nascar == "the south"

why would the midwest specifically be a problem?

Outside the Bay, Ann Arbor & Detroit are the leading cities conducting research on autonomous vehicles thanks to the University of Michigan and the motor industry.

Besides most people have 45 min commutes. If your car came preheated each morning and drove you to work while you take a nap, then people would love it.

> region proposal network (RPN)

sorry RPN is already taken !

Do you mean: Risk Priority Number? Reverse Polish Notation? Registered Psychiatric Nurse? Radio Philippines Network?

Surely you're aware that acronyms are sometimes domain specific and have multiple meanings.

Reply on news.ycombinator.com