Hacker News Reader

Comment by seibelj | Hacker News Reader

I truly believe that general purpose self driving cars are much farther away than the latest press releases would suggest. Apparently they don't work well at night, in the rain, when the road lines aren't perfect, etc[1]. They basically only work in the perfect weather and roads of Mountain View during the day.

In Boston, the roads are shit, the drivers are crazy (including me), and everything is permanently under construction. Once people learn how to spot these cars, very quickly the tricks to passing them are going to be learned. Not to mention the jokers using laser pointers and whatnot to mess with them. If you think this won't happen, you are sorely naive.

I think self driving cars will work for specialized cases, like long haul trucking on highways, but for the commuter, it's going to be a long ways off.

[1] http://gizmodo.com/6-simple-things-googles-self-driving-car-...

View on HN

The first time an airplane flew it only went 120 feet (37 m).

11 years later the world had it's first regularly-scheduled commercial airline. [2]

A lot can happen in 10 years.

[1] https://en.wikipedia.org/wiki/Wright_brothers#Flights

[2] http://www.space.com/16657-worlds-first-commercial-airline-t...

I think it's more amazing we went from 0.02 miles one way, to 477800 round trip to the moon in just 66 years.

I vaguely recall the google director saying their autonomous cars already drive better than his parents.

Even if it's only daytime clear weather driving, super cruise control would be insanely useful for commuters, truck drivers, and city busses.

It's easy to fall into the trap of perfect or nothing. Even partial solutions create a lot of value. Complete autonomous driving will be here real soon, but there might be a few intermediate steps along the way.

> super cruise control would be insanely useful for commuters, truck drivers, and city busses.

There’s something that does exactly that, is cheaper and safer: It’s called rail.

Does rail go straight to everyone's office building?

It doesn’t have to. In a properly built city, the rail will end up less than 500m from your destination, and you can walk the rest.

Example cities with okay or good rail infrastructure: New York, London, Paris, Vienna.

You also save a LOT of time, because you are never stuck in traffic, and you can use the time in the train to check up on your emails, etc – and you never risk getting into an accident because you are tired after work.

I don't live in a "properly" built city. Do you?

I live in a european city founded around 1200 – so, while we destroyed our tram system, and our public transport is shit, I can still get everywhere with public transport equally fast as driving, often faster.

I’d like to have even better infrastructure, and 24/7 service, but I don’t have an issue. I have 3 bus stations within of 300 meters, and a train station 800m from my current home.

> It doesn’t have to. In a properly built city, the rail will end up less than 500m from your destination, and you can walk the rest.

Part of the problem in 2015 is that unless there was planning for rail from the start, the cost to build out rail in established areas is enormous - typically because you need to tunnel - in addition to the cost of community consultation and other access-related issues.

I otherwise agree that well-planned rail into key areas would have benefited a lot of cities, especially those feeling the pain from road congestion now.

Well, you don’t always need to tunnel. You can start by adding light rail to the streets.

Make every lane 30cm thinner, remove 1.5 lanes, and you have enough space for a bike path on every side and light rail tracks.

"You are never stuck in traffic"

Clearly you've never ridden the subway in NYC during rush hour. Most of the major lines will have some delays, you'll have to wait for multiple trains to pass before you can get on, or once you're on the train will stop because "we are waiting for train traffic ahead".

Well, as I said, MTA's service is more on the "okay" and not on the "good" side.

I’ve seen bad train traffic, too (here is my city at 2am during a festival http://i.imgur.com/KyWkUzJ.jpg), but still:

Imagine the same amount of people using the train now being on the street, everyone in a seperate car.

I commuted by tube in London for 3 years, my most recent house was 1 minute walk from the station and my office was also 1 min walk from a very busy central station. So basically door to door. Even still, commuting was not a pleasant experience.

The human congestion in the central tube stations at rush hour is appalling, as is the air quality down there, as is the lack of personal space on the train itself. It's a very stressful, undignified experience for everyone.

I'd much rather be spending a bit of extra time in traffic, sat comfortably in a self-driving taxi, than take the tube. That said, I think there's good reason to believe that as self-driving cars become more prevalent, traffic flows will actually improve due to better flow management, less need for traffic lights, and fewer cars parked.

Rail does the job, but it does not do it elegantly. It's very unpleasant to take mass transit railways in major world cities. Nobody who's done it for any length of time would reasonably contest this. Self-driving personal vehicles could be a major improvement on the commuting experience, if indeed the future involves commuting at all - a vast increase in remote work may simply sidestep the issue completely by the time self-driving cars are a mature tech.

The issue is that there’s not physically enough space for self-driving taxis for everyone, not even if you compress them down to Messerschmitt Kabinenrollers. http://www.autoplenum.de/Bilder/testreports/jwo2009050316403...

A single person in the tube takes, technically, 0 space due to being in a tunnel, and even in overground railway a single person takes less than a square meter.

With a car, you waste far more space.

And, by the way, I’m taking public transport every day, all year round, even in situations like http://i.imgur.com/KyWkUzJ.jpg and I still find it far more pleasant to wait 20 minutes in a crowd than to be stuck for 40 minutes on the highway.

> there’s not physically enough space for self-driving taxis for everyone

Do you have any sources on this? I agree that space on the roads of major cities is an issue now, but several changes in the future could mitigate these issues significantly:

1) Self driving cars will probably end parking in city centres. Suddenly, a lot of congested roads have ~twice the surface area.

2) Self driving cars may eventually be able to drive within centimetres of one another.

3) No need for a huge bonnet (small electric engine), boot/trunk (rooftop rack storage), pedals, dashboard, etc in an electric automated taxi. A 2-4 person vehicle is now rather compact indeed.

4) Remote/part-time-remote work may mean commutes are avoided, or staggered throughout the day, for the majority of people. There's a lot less peak capacity required, compared to today.

Well, your car will still have only one person – and you won’t get it smaller than the Messerschmitt Kabinenroller I linked above. And even that is more than twice what a person in an overground train takes in space, underground trains are essentially free.

Additionally, in many cities there is already almost no downtown parking.

It’s not like you’re going to turn the laws of physics over.

One of the sources why cars are worse than bikes or busses, from the city of Münster: http://media.treehugger.com/assets/images/2014/07/Public_Tra...

Safer, perhaps. Cheaper? Not so much. Rail infrastructure is expensive, both to build and to maintain. Rail is also much less tolerant to slopes and tight corners than even the most unwieldy of road vehicles (which contributes substantially to the expense; lots of terrain engineering involved in order to build a railroad that's actually safe to travel on).

That said, rail does have superior cargo and passenger capacity, so these costs can be offset by the increase in potential revenue.

Rail is, compared to the huge amount of highways and interstates built, quite cheap. Especially if one wants to expand capacity of highways to counter population increase.

The sheer capacity of rail makes it even on small scale sustainable, on state scale it’s even more effective. (especially on the east coast).

And with cars having such low efficiency (transporting 2 tons of metal for a single passenger using gasoline at 90km/h seems a waste when you can have a single train transporting 400 people at 320km/h using electricity instead) it’s just not really effective on large scale.

New York City is a good example in the US, but cities in Europe and Japan show how even small cities and whole countries can benefit from rail infrastructure.

By "Japan" you mean major cities in Japan. An hour or two outside Tokyo, public transportation is 1 bus every 30 minutes. A friend and I were in Yamanashi-ken last year, and we were complaining to a random shop owner about the bus schedule. She said, "twice an hour is amazing! this isn't Tokyo, you know!"

Most people out there drive if they have to go somewhere.

Twice an hour is actually not that bad – I live in a suburb of a German city with the bus to my uni only coming every 30 minutes, and it’s not a problem – almost everyone uses the bus or bike. You just have to get used to the thought that not everything happens when you need it, but instead you have to adapt to when things happen.

And people often use that, the NYC subway is a great example. But it's not so effective in lower density, or less established rail systems. You'll often see east/west travel relatively accessible, but north/south hard, for example. Rail only really serves those along the line. I'm sure phoenix could build a ring and spoke system to connect up all of those low density developments.

The thing that sucks though, it's a big upfront cost. super cruise control is an incremental cost that takes advantage of existing infrastructure.

Rail isn't so great at getting food to grocery stores. (cities, sure, but not actual transportation endpoints.) Such a system could be built, but trucks are so obviously effective, and they work with the existing infrastructure.

Rail is cool but a big upfront cost for something that's not obviously superior to the current system seems wasteful.

Maybe fill train stations with self driving taxis to cover those last few miles to home, or work, or wherever.

Yes, the thing is just that the interstate system was in the end more expensive than similar rail infrastructure would have been (especially in maintenance).

And, well, yes, cars and trucks will still be necessary, but not nearly in the same amount.

I'd love to believe this quote, but light rail per mile is an order of magnitude more expensive than highway per mile. Do you have any citations I can look into?

Thanks.

The issues are essentially based on (a) how many lanes your highway has, (b) how many complex interchanges you build (interchanges with many bridges tend to lead to a cost explosion), and (c) rail tracks are usually able to be used for decades without major maintenance, while highways usually need to get filled and fixed multiple times a year.

I only have a specific study for my city, where they were comparing the costs from fixing the streets in the cities from the busses running over them vs. building light rail instead, and in that study it turned out the light rail would be, if you also consider stuff like maintenance and accidents, cheaper.

I can't find anything more formal with a quick search, but this is pretty plausible http://alankandel.scienceblog.com/2014/01/11/rails-vs-roads-...

$35million/mile for rail, $25million/mile for a single lane. Rail maintenance is cheaper, and ongoing costs are generally lower - you don't need police patrolling railroads.

Railroads are cool, and they have a lot of advantages, but we've got this huge investment in our existing system. And unfortunately, unlike self driving cars, there isn't a smooth transition. It's like switching from pc to mac, but all of your software is home made, and must be rewritten. You have a huge upfront investment to just get started, and you have to maintain the other system as well.

The two key advantages of road over rail:

- much shorter stopping distances. This lets roads be "general access" in a way that rails have never been, even if you were to have a railway running past your house.

- much smaller minimum turn radius

- ability to drive around obstacles

The situations where roads are better:

- low amounts of people (or goods) - going into different directions - coming from different places

The situations where rails are better:

- high amounts of people (or goods) - going into the same general direction - coming from the same general area

Which is why roads are damn useful in rural cases, but if you think about building a 23-lane highway, maybe rail might be better.

Most of the places I drive to don't have trains going there. Sure its great between cities, but getting into the mountains?

The point of the parent was that for commuters, for long-distance cargo transport and for busses some "super lane guide" would be useful.

Does your going into the mountains contain "driving on the same lane with thousands of others from the same place to the same place every day"?

Because that’s what I was suggesting – nothing more, nothing less: Replacing mass-scale commuting with rail.

Yes, if you drive somewhere where no one else wants to go, cars are still useful – but if you’re going the same path that another million people are also taking, maybe rail might be better.

Rail and Road are not exclusive, they can coexist.

But how old are the guy's parents? Are they still alive?

Not the article with the quote, but a similar spirit. http://www.extremetech.com/extreme/187438-googles-autonomous...

Sure, they're limited, can't drive in snow. Snow shut down Atlanta because humans couldn't hack it. Autonomous cars can't compete with humans at their best, but they're consistent and know their limits.

It's tough to see in tech, but some older people's driving fades fast. They arrange their lives to avoid driving on the freeway or at night.

Snow shut down Atlanta because the city did no preparation on the streets and didn't salt them, so the whole city was covered in an inch of ice. I don't think any car could drive on that as long as it used tires.

There were also (according to local legend) three snowplows for the whole city in the 2000s.

Is your analogy useful? It's been several decades since the first driverless cars, so if the "Wright Brothers to St. Petersburg-Tampa Airboat Line" comparison were predictive, we should have had plenty of those cars on the road by now.

There are also many other analogies you could have made with different timelines. Virgin Galactic, which bills itself as the first commercial spaceline, is already half-a-decade behind its original launch date. Even Tito's 2001 flight was four decades after Gagarin.

Why look to commercial flight and not commercial space travel as the relevant comparison?

> Once people learn how to spot these cars, very quickly the tricks to passing them are going to be learned

Crazy driving is itself often tricks to pass human-driven cars.

> Not to mention the jokers using laser pointers and whatnot to mess with them. If you think this won't happen, you are sorely naive.

No need to wait; laser pointers are effective at messing up human vision sensors today. But no epidemic yet...

Most of the operation of current society depends on the vast vast majority of us not being in a mass-murderous mood.

It's harder to be aggressive when dealing with human-driven cars because people can react. The commuter dozing in his self-driving car is not aggressively driving, and with his car legally mandated to be in super-safe mode, it's going to take twice as long for it to make it through rush hour traffic.

> it's going to take twice as long for it to make it through rush hour traffic.

So read a book or work on something on your laptop or go on Amazon and buy a present for your partner with the health insurance money you'll eventually save by registering your commute as autonomous.

You're using a first approximation to a self-driving car ("follow all laws to the letter") to extrapolate.

There are a large number of reasons you'd notice that won't work the very first time you take a car out for a drive (you give a great one: "it's going to take twice as long for it to make it through rush hour traffic"). With no eye contact or hand gestures available it's a harder problem, but aggression and counter-aggression is going to have to be a part of any self-driving model. Even if all cars were self driven you'd still have pedestrians willing to cross the street at any sign of hesitation.

Whose problem is this? If I am in a self-driving car, I don't expect it to use "aggression" to try to compete with other drivers. I just want it to get there safely.

The robot equivalent of a "work-to-rule" strike could be a possible response to this kind of behaviour: the driverless cars all around an aggressive driver could go into a slower, more defensive mode and box the aggressive driver in—to ensure the safety of neighbouring traffic, of course.

That's a horrible idea. What if the aggressive driver is rushing a bleeding passenger to the hospital?

Even in your made up thought experiment that accounts for << 1% of all aggressive driving miles I still think this is the correct approach to take. Speeding to the hospital is only superficially different than speeding and weaving through traffic on your morning commute. The biggest difference is probably that you are flustered, and less likely to make good decisions while in route. The worst accident I have ever seen was someone rushing to the hospital [1]. The threat to those around you, and the chances of an accident slowing you down outweigh the small potential time saved. It would be far better for you to learn first aid and respond appropriately than for you to drive aggressively to the hospital.

[1] So there I was[2] driving through Iraq when an Iraqi army convoy approached rapidly from behind us. While swerving past my trucks, that had pulled over for them to pass, one of the vehicles left the road and rolled over multiple times. 2 people died in that accident, I applied tourniquets to 2 additional people who lost limbs, and no, they didn't get the gunshot victim to the hospital any quicker that way.

[2] Fun fact. Do you know the difference between a fairy-tale and a war story? War stories start with "So there I was".

Good point. And what about police/emergency vehicles rushing somewhere?

Emergency vehicles rushing around may or may not be good overall. They have a few advantages over your average crazy driver though such as lights and sirens to get the other vehicles attention so they can move out of the way.

Police often have additional driver's training. I happen to think that police often drive faster than they should, but this is less the case for emergency vehicles. I rarely see an ambulance or fire truck moving at high speed. They are pretty aware of their job to get somewhere safely, and I am typically only passed by an ambulance if I am stuck in traffic myself.

Traffic police will be looking to make up lost revenue from being unable to ticket autonomous car riders, so those aggressive drivers will be a better target. (Not that I approve of using tickets as revenue, just saying that it will happen.) And those autonomous car passengers now have their hands free to take video of the unsafe human drivers nearby, for later public judgement ;)

I've been thinking about this and cringing in wait for the lobbying against self-driving cars from police unions.

But then I thought, that couldn't happen, because it's too clear that any such lobbying puts police unions squarely at odds with citizen safety. Now, shoot-first cops, civil forfeiture, and highly militarized police forces already point at this, but there's plausible deniability. Opposing self-driving cars could be too direct of a swipe at public safety from public safety officers.

I don't know...

I agree. Law enforcement will definitely have to change, though. Right now, traffic infractions are a way for the police to look for other criminal activity. Pulled over for speeding? Cool - let's run your license to see if you have any warrants. Let's see if I can find something else that you're doing.

Self-driving cars will completely remove this avenue of investigation because there will be much fewer infractions.

I think that this is a good thing and a bad thing. On the one hand, you won't get police harassing random innocent people on the off-chance that they're criminals. On the other hand, the police will start doing other things because they don't have to pull people over anymore, and those things might be much more invasive.

If you're not as white as I am, it's 99% good. Driving while dark-skinned is a big cause of traffic stops.

Rush hour traffic on a highway is a good example of where aggressive driving doesn't actually work very well. I've seen people try to drive aggressively (switching lanes and tail-gating), and they're still in sight for a very long time. So they're maybe seconds ahead of me? Sometimes I even end up passing them because they chose the wrong lane, or they ended up stopped at a light.

You do get idiots saloming around in somewhat heavy traffic, but that doesn't actually slow others down much while they're putting everyone at risk.

I'm skeptical of that. I think on average drive time goes down. A few steady drivers drastically reduce stop and go driving, and can eliminate those weird traffic aftershocks where everyone slows down for nothing, just because someone slowed down at that point an hour ago. [1]

Furthermore, the total time might be less on a good day, with aggressive driving, but the more aggressive drivers there are, the higher the likelihood of an accident, which takes substantially longer than slow drivers.

t1 = aggro drive time t2 = slow drive time t3 = accident drive time

small cities, say less that 500k, it's probably not worth it, because p(accident) is pretty low, freeway backups relatively rare, say in the neighborhood of .1

bigger cities though, traffic gets messed up all the time. t2 wins.

t2 < p(accident) * t3 + p(1-accident) * t1

http://www.npr.org/2013/11/29/247825768/phantom-traffic-jams...

Rush hour traffic is going so slow that non-aggressive driving will only cost you a low single digit percentage of travel time, which will all be made up at the first red light.

I routinely drive at 90km/h on the main 100km/h stretch of my 28km commute. 23km at 90km/h, and I pull up next to or immediately behind the person that passed me a few minutes ago: the difference in arrival time between a car doing 120 and another doing 90 on that stretch is 4 minutes, assuming they hit 120km/h as soon as they leave the lights at one end of the road. If they have to overtake slower traffic like busses, cranes and me, they get less of a head start.

Eleven minutes versus fifteen on the only stretch where aggression makes a difference. There is no way in the world that this counts as "double" the time. The rest of the trip is the same average speed of about 25km/h, with the more aggressive driver getting to use their brakes more than me.

Red lights are the great equaliser.

Aggressive driving buys you one or two cycles of the traffic lights, along with a much higher fuel and maintenance bill.

> The commuter dozing in his self-driving car is not aggressively driving, and with his car legally mandated to be in super-safe mode, it's going to take twice as long for it to make it through rush hour traffic.

That's a common idea based on not-very-careful observation and just isn't factual. There's plenty of data showing that aggressive driving has minimal effect on travel times. In fact, you actually get a bigger speed-up by strategically avoiding left turns: companies with delivery fleets plan their routes with this in mind.

[1] Not-super-technical explanation: http://lifehacker.com/does-speeding-really-get-you-there-any...

As well as those points, it also seems that the on-board sensors for the car are not good enough, as Google has spent a great deal of time scanning the roadways and constructing high resolution models of the environment. The cars apparently use this pre-prepared data to help them drive well.

This strongly suggests that the cars, in their current form, are not as autonomous and self-reliant as people assume. It also implies that the cars are not going to cope well if they encounter sudden changes in the environment.

Does it matter if they rely on some pre-processed data, as long as they have some form of fallback that allows for manoeuvring when a change in environment is found?

Human drivers also face immense difficulty when there is a giant tree fallen across the road. Luckily, such changes in environment are relatively infrequent.

Google Maps’ data about my city are from 2005. Half of my district is not even existing – and with Google Mapmaker not available in my region (northern Germany), I don’t think it will be added anytime soon.

While currently I end up with dozens of human drivers every day stopping in front of a retractable fence (and some try to trick it), do I have to expect Google’s cars soon just driving right through it, or refusing to use alternative routes?

Especially because we have shit weather with lots of wind and rain and snow and fog and below 10m vision often during winter. How’s the Google car supposed to drive in that situation?

Slowly, just like a human?

The issue is that the reason the fence is retractable is because of the busses that speed through there every 7 minutes. If you drive there slowly, you might cause a crash. Cars shouldn’t drive there at all, but Google Maps still suggests to drive right through it as the fastest route. And it’s just one example, we have hundreds of them all over the city.

If they can’t even provide reliable maps, how can we expect reliable self-driving cars?

How do humans learn to not go through the gates?

I'm assuming that self-driving cars will also perform data collection for the rest of Google's self-driving cars. I could be incorrect about that, but if they do, it won't matter that the current data is from 2005. Alternatively, if Google do need to collect manually, I would assume they will mobilize to do just that. They did it for Street View, a feature that isn't directly linked to revenue. A much more solid business case could be made for doing it for self-driving cars.

Humans rely almost solely on vision from a single source (counting both eyes as just one source) when driving. Autonomous vehicles have far greater potential when it comes to sensors. Google's current vehicles utilize LIDAR and GPS. Tesla's lane following uses cameras, but they get more than one perspective. If it added anything, infrared could be used. There's potential to have tonnes of cameras on the car to get a fuller view. They can triangulate position more effectively by listening out for Wi-Fi and cellular network signals. There's gyroscopes, too. CPUs can calculate exactly how much stopping distance is required in the particular road conditions and adjust speed accordingly. When you think about the sensors and processing capability compared with two human eyes and a brain, it seems to me as though humans are very poorly equipped for dealing with driving in hazardous weather.

You're probably right that the current cars struggle under poor weather conditions, but it hardly seems like an unsolvable problem. Once they do function in poor weather conditions, they'll probably do a far better job of it than a human driver would.

Well, Google doesn’t even have StreetView imagery of most of Germany after they intentionally broke the law to get better images¹, so it’s not expected they’ll do much better anyway. I talked today with a bunch of friends, and we discovered that for all purposes here maps works better in Germany and Austria (except in Hamburg, that actually works in Google Maps. Vienna's satellite view is unusable in Google Maps)

1: To take images from just above 2.50m height you need permission from every house owner, cause most house owners build their hedges and fences in expectation for a person with an eyeheight of 1.90m or less. Google actually intentionally increased the height after their first test runs because they were annoyed with hedges blocking the sight. According to the news back then.

---

To your point: Well, the issue is, if Google is not willing to provide or collect up-to-date data – even if they get it for free – then data-based driving is quite risky.

And in the narrow, crowded streets of european cities I’d rather not have an automated car drive on sight in a winter storm.

> Once people learn how to spot these cars, very quickly the tricks to passing them are going to be learned.

You say that as if arriving at your destination one car-length in front of a self-driving car somehow renders autonomous vehicles worthless.

Who cares if they stop or slow down when you cut them off? It's only idiot humans who think that the reward pathway signal of "woohoo I'm in front" means that you've won something. You haven't.

Reminds me of that Mythbusters. The junior team split in two, one drove aggressively over-taking, switching lanes constantly, while the other stayed in a single lane, didn't overtake, and just arrived as traffic allowed.

They arrived within a minute of another another, and the aggressive driver (who was Grant I believe) reported that it totally wasn't worth all the stress to arrive so close.

People who stress, get upset, and try to maximum their arrival time down to a second are only hurting themselves.

Aside from dangerous driving I try to keep as calm driving as possible, and like to just get a safe distance (4 sec rule) behind a semi on the freeway, and sit on cruise control until I arrive. Even if the space I give isn't massive, a lot of people utilise the space between me and the semi to transition from the fast lane[s] down to the slow lanes/exit, which I think is a legit public service, nothing worse than getting stuck on the freeway because you cannot transition.

I use "driver assist" in my car every day, for about 50 miles. It keeps my car in the a lane, follows traffic, and lets me concentrate on my podcast.

It's not perfect, I sometimes have to step in; but it works at night, it works in all but the heaviest rain, it works in snow... and this isn't some crazy future product, it's a thing I already own.

Self-Driving cars are coming faster than the naysayers expect.

"I sometimes have to step in"

Yes, those are exactly the difficult edge cases that are going to make the reality harder than it may seem.

I feel very much the same way. A significant portion of drivers also deal with winter/icy conditions for at least part of the year, and I have a really hard time believing that self-driving cars will be able to handle those conditions well, at least not for a long time. You have snow falling, snow covering many/most of the lines on the road, varying degrees of traction - added to the uncertainties of pedestrians, construction, other crazy drivers, etc. Heck, the sensors alone are going to have an inch of snow and ice on them after being outdoors for a couple of hours. Winter driving is hard enough for highly-focused human drivers, making decisions that are often based on factors that don't easily translate into algorithms and computer models.

Google speaks of a need to leapfrog the driver assistance systems and instead focus on fully-automated driving systems because the driver assistance technologies won't incrementally improve to the level we need for full driver-less cars[1]. I think that's a fair point, but I'd go one step further: self-driving cars thrown into a mix of human-operated vehicles and roads that aren't specifically designed for self-driving cars is (in my opinion) a similar mistake. I believe we'll reach a point very soon where great technical advances in the vehicle alone will have diminishing returns in terms of safety on the road.

If we are going to solve this problem once and for all, I'd be much more enthusiastic about a self-driving system that operates only on distinct roads/lanes/tracks that are designed only for self-driving vehicles. Perhaps that's an unreasonable expectation, but trying to design systems that anticipate and make decisions based on the (often irrational) behaviors of other human drivers seems no different than designing a full human-level artificial intelligence - and even then, it'll still be prone to mistakes because other humans are involved.

Either we get to the point where all vehicles on public roads are networked and cooperating through software to self-drive, or we're just making guesses about other people's behaviors. As Chris Urmson puts it, "just making the cars [and I'll add, roads] incrementally smarter, we're probably not going to see the wins we really need."

[1] https://www.ted.com/talks/chris_urmson_how_a_driverless_car_...

Remember though that we're coming from a position where humans sometimes don't drive very well while sober and in daylight, not to mention night, rain, while drunk, tired, etc. We're often selfish or distracted. An automated or augmented car with cameras can see virtually instantly in all directions.

When I'm driving, my rear view is somewhat obstructed by a toddler's car seat in the middle of the back row, my blindspot is further obstructed by a sunblocking mesh on the infant's side. When I check my mirrors or especially the blindspot, I am taking my eyes off the road in front of me.

I think I'd rather a superior robotic driver that stalls in tricky situations than all of us crazy humans making panicked decisions when facing pressure.

Google doesn't need to fix all those problems in order for self-driving cars to be in widespread use. Even a self-driving car that works only during daytime would be very successful.

> "when the road lines aren't perfect"

Ok these aren't coming to Montreal for a long time!

Reply on news.ycombinator.com