How Google's Autonomous Car Passed the First U.S. State Self-Driving Test | Hacker News Reader

How Google's Autonomous Car Passed the First U.S. State Self-Driving Test

spectrum.ieee.org

64 pointsspectruman55 comments

I really feel autonomous cars are a prime candidate for Next Big Thing over the coming 10-25 years.

Combined with transportation platforms such as Uber, I wouldnt be surprised to see fleets of robotic taxis exclusively servicing dense urban areas in the next 15 years. Its one of those great problems that can scale locally if an enthusiastic city council or government get behind it. Something like an extension of what London has done with their congestion charge: http://en.wikipedia.org/wiki/London_congestion_charge

Just imagine the transition to an inner city driverless fleet. No more parking problems in metro areas. Massive swathes of real estate become available as parking bays are no longer required. Markedly faster travel from A->B as networked cars work together to ensure optimal traffic flow. Once all the meatspace human drivers are off the grid, you can remove traffic lights.

Australia is well placed from a rural transportation perspective - we have many small towns 500km apart and plenty of road freight. Accidents on these roads are usually from driver fatigue, speeding or hitting wildlife. These roads are 'easy' for driverless cards - long and straight, not many surprises (except the odd kangaroo which I think the cold heart of an AI would handle better - the worst thing you can do is swerve to avoid them and roll your vehicle).

I think when the kids of 2150 sit down in history class to study the 20th century, they'll laugh at the idea that every family invested significant money into 1-2 expensive hunks of steel for transportation. We fling these big dirty hunks of steel around at 100mph for perhaps 5% of the day. It can get quite dangerous! Then we waste huge amounts of land by storing them for the other 95% of the day. How bizarre is that?

> I think when the kids of 2150 sit down in history class to study the 20th century, they'll laugh at the idea that every family invested significant money into 1-2 expensive hunks of steel for transportation. We fling these big dirty hunks of steel around at 100mph for perhaps 5% of the day. It can get quite dangerous! Then we waste huge amounts of land by storing them for the other 95% of the day. How bizarre is that?

Considering folks in rural areas will still need personal transportation in 2150 I don't think it would appear bizarre at all. Do you consider it bizarre that people used to keep horses around (which means stables, workers to care for them, feeding them, etc.) just for the occasions where they'd want to go into town? And even with self driving cars does that suddenly mean I'm prohibited from owning my own so I can travel where I want when I want? I really don't understand this love affair people have with the idea that we should let some corporate computer network decide when and if we're allowed to leave our homes.

I think you miss the point and are being overly semantic.

All I'm trying to convey is my feeling (shared by many) that autonomous vehicles represent a technological step change that will change the way we live. Similar to how the internal combustion engine, radio, antibiotics, transistors, telephones and the internet have changed the way we live.

Do you think a child today would consider it bizarre and surprising that in 1850, if you got the flu, you'd probably die and it might take 3-4 days of travelling to arrive at a medical centre?

>I really don't understand this love affair people have with the idea that we should let some corporate computer network decide when and if we're allowed to leave our homes.

I should have read this first, you're just being silly. No one is suggesting this is the case.

s/semantic/pedantic/

I know, that's like the worst correction ever. I couldn't resist. b^)

> And even with self driving cars does that suddenly mean I'm prohibited from owning my own so I can travel where I want when I want?

On a private road sure, knock yourself out. But by driving on a public road you're risking the lives of yourself and other participants (and it's not a "everything is unsafe" rhetoric, cars kill a lot of humans every year), and the whole point of introducing self-driving cars is to get rid of human drivers on public roads because they kill people.

> I really don't understand this love affair people have with the idea that we should let some corporate computer network decide when and if we're allowed to leave our homes.

This "love affair" comes from embracing the fact that a centralized algorithm can route you much faster, safer and more efficiently that you could ever do it yourself, therefore it is best to trust the machine. Because the machine is really better at this than humans.

Machines are much, much better at flying airplanes too; yet all commercial planes use pilots. I wonder which one will change first, planes or automobiles, which has the most resistance to change. Airplanes - decades of resistance now. And flying is a trivial problem compared to driving.

I'm not an aviator, but I was under the impression that commercial flights fly on auto-pilot for 99% of the trip, with an exception for take-off and landing. I don't know the reasons why humans handle lift-off and touchdown though.

Still, commercial airplanes are much, much safer than cars, therefore I'd expect a push for self-driving cars to be stronger than for self-flying planes.

And where are all the flying accidents? Takeoff and landing, and mostly landing. A case of human cussedness overruling logic. The autopilots can land a plane in all conditions and respond to sudden changes with lightning reflexes.

Some of it probably boils down to people still not trusting machines. I guess this will change within a generation or two, as the kids growing up in increasingly automated world mature.

Perhaps, but there's a big difference between not trusting machines and not trusting them to be perfect. If people don't trust machines, it's because they're surrounded by machines that break and malfunction constantly, systems that don't work, networks that fail - and autonomous cars implies a level of faith in all of that working flawlessly most of the time in spite of demonstrable reality.

While I object to autonomous cars mostly because they remove human autonomy and implicitly limit freedom (as a tradeoff for safety and greater efficiency), I think it would be a bad idea in any case to come to the point where having a steering wheel isn't even an option.

> If people don't trust machines, it's because they're surrounded by machines that break and malfunction constantly, systems that don't work, networks that fail - and autonomous cars implies a level of faith in all of that working flawlessly most of the time in spite of demonstrable reality.

People are falling to a very common bias. They see machines that break and malfunction constantly because those are the only ones noticable. They don't see many more machines working flawlessly, even though their very lives depend on them. Auto-pilots on planes, HVAC automation, gas delivery automation, lifts, food processing machines, electric grid, the Internet - all of this works almost perfectly, but we notice only occasional hickups and failures. The demonstrable reality is that most of the time things really do work flawlessly.

Still, it doesn't matter. People need to understand that it doesn't matter if the machine isn't perfect. What matters is if it is better than human. And for self-driving cars, outperforming human drivers isn't a distant goal.

> While I object to autonomous cars mostly because they remove human autonomy and implicitly limit freedom (as a tradeoff for safety and greater efficiency), I think it would be a bad idea in any case to come to the point where having a steering wheel isn't even an option.

Well, but that's the exact point of having self-driving cars. Because humans should not have the "freedom" of driving around on public roads and killing each other all the time. Or, in other words, as a proponent of self-driving cars I want to be able to exercise my freedom of not getting killed by myself or another driver, because human reflexes and cognitive skills are not up to par with the requirements of driving a motor vehicle.

Public roads are infrastructure, its goal is to let people move from point A to point B. The requirement that people do the moving themselves is only there because we didn't have self-driving cars then, and cars are more flexible than trains. Sure, let's drive all we want on private property. Have our freedom. But the infrastructure works better when humans don't mess with it.

You make a good point, but I wonder if autonomous car proponents aren't falling for the same bias in regard to human drivers.

It may be the case that self-driving cars are safe. However most of the time, most human drivers don't kill each other all the time, and the majority of driving events are uneventful. It is a risky activity but it's certainly not the bloodsport some make it out to be.

You're right. Of course majority of driving is uneventful and the activity is not a bloodbath. Your average driver isn't a speed maniac or someone who loses control of the vehicle three times a day (though an average driver I meet has a bit, let's say, relaxed approach wrt. speed limits and safe driving distance).

But because driving is something that a great many people have to do, often many times a day, it still sums up to a significant death toll worldwide. It's a statistical thing, not emotional one (no one is honestly impressed by traffic accident these days; bomb on a plane on the other hand...). By introducing self-driving cars we can significantly reduce if not outright eliminate those millions of unneccessary deaths while also improving energy usage and efficiency of transportation as a whole. In other words, no traffic jams, you get quicker everywhere and the world as a whole uses less energy to move people around.

I think you're misunderstanding, icehawk219 was talking about individuals owning their own self-driving cars. That's a separate question from allowing human drivers on the road.

You're right, I misunderstood him. To answer that point then; I don't think people will be prohibited from owning a self-driving car, not initially; I believe they'll relinquish this privilege because of practical reasons - not having to store, maintain and protect a car is better than having to do it. You can see this happening in big cities - people decide not to own a car because it's too big of a mess without real benefits to them; it's more economical to hop into a bus or even get a cab. With fully networking self-driving cars we should expect to see many improvements in efficiency of shared transportation, further reducing the need for individuals to own a car.

I'm worried about the implications on the labor market. Here are some quick numbers from the BLS about US employment: There are 3.664 million motor vehicle operators with an annual mean income of $35,000.

That includes 158,000 transit bus drivers, 500,000 school bus drivers, 170,000 taxi drivers and chauffeurs, 1.6 million tractor-trailer drivers, 777,000 delivery truck drivers and 400,000 driver/sales workers (restaurant take out, newspaper delivery, etc)

By comparison, there are 3.7 million working in all computer and mathematical occupations.

[http://www.bls.gov/oes/current/oes_nat.htm#53-0000]

I believe that there's many many more jobs at risk but it doesn't need to be a bad thing. Once a robot can accept orders and cook a burger as well as a McDonalds employee, won't we lose them too? Many many jobs will be automated.

As a society we need to find a way to handle this so that people out of work have their needs met and aren't just left to starve but I'm not sure we can expect to keep employment so high.

This might be the single thing I find most absurd and frustrating about our current society. We've somehow arranged the incentives so that it's a very bad thing to reduce the amount of work we need to do to maintain our standard of living.

As robotics and other technologies continue to improve, we're going to keep hitting this problem in a large number of fields.

What happened to all the typewriter repairmen when computers became ubiquitous?

Certainly there will be upheaval in the labor market. I can certainly see unions and lobby groups pushing back against autonomous vehicles and delaying adoption.

But I dont think its significantly different to other technological revolutions (the wheel, the transistor, the computer, the internet).

> Once all the meatspace human drivers are off the grid, you can remove traffic lights.

I'm not sure you want to do that. In city centres, the traffic lights are there to stop up pedestrians temporarily so cars can get through.

> Once all the meatspace human drivers are off the grid, you can remove traffic lights.

As a driver, I've silently tolerated horse-drawn carriages, pedestrians, cyclists, motorbikes, tractors, etc on the roads. And I expect autonomous cars to extend the same courtesy to us drivers. If there's a situation where autonomous cars are incompatible with human drivers, then that situation will have been deliberately created. Human drivers have paid the taxes that built the roadways, in part directly through gasoline taxes, and it would be unfair for large corporations to re-appropriate public infrastructure to further their profits.

The passengers of both types of cars are the same, the only difference is that one doesn't have a subset of passengers also drive.

Autonomous cars will only really see true potential when they don't need a driver at all... or one in the car. Think one driver, jumping in on 5 or 10 UPS delivery vehicles as they park and launch delivery drones locally, or 20 Google maps vehicles with one remote driver jumping in at roundabouts. While 3 city buses with capacity for 5 or 10 more people each since there's no physical cockpit is promising or not having to pay attention, the true potential is all the things you don't need people for.

The true potential in Autonomous vehicle removes the cost of having people involved with delivery, exploration, etc.

Once the tech is there, you can remove the excess weight and volume of seating, safety mechanisms (mirrors, belts, roll bars, airbags...).

Removing human ergonomics will also allow for more efficient designs per purpose.

> Once all the meatspace human drivers are off the grid, you can remove traffic lights.

You will always have humans on roads, whether they are pedestrians or cyclists. Not everyone is going to want to drive everywhere - if nothing else than for cost and exercise reasons.

You can't turn roads into people-free death traps. In the UK, the law is very clear that cyclist are permitted to cycle on a road (that isn't a motorway), even if there is a parallel cycle path. However, many people take the view that a cyclist "should not" be on a dangerous road, and that if an accident occurs then it is the cyclist's fault for being in the wrong place. There is quite strong pressure from cycling groups to keep roads safe, rather than banning cyclists from them.

If the cars can sense you and stop in time you don't need to remove the cyclists and pedestrians from the road. Having no misbehaving driver would arguably be safer for the cyclists (until a firmware update crushes them all).

That works if there is only one cyclist or pedestrian on the road. Traffic lights are required when there are multiple, so they don't bump into each other, regardless of whether the self-driven cars can avoid them.

Traffic lights aren't even necessary for autos, if all drivers are minimally aware of their surroundings. (Admittedly that situation is uncommon.) It would be extreme authoritarian overkill to use them in a bicycle-only situation. Appropriately-placed "yield" signs are quite sufficient.

Like it. But automatic cars is about more than safety - its about getting people places quickly. Cyclists and pedestrians on the road would drive traffic speeds down to nil, especially if safety trumps all.

The car may be programmed to adhere to strict safety laws, but some necessary maneuvers will not be possible due to physical constraints and people can still get hurt.

I find it interesting that google is building a self-driving car for public roads. There's such a vide variety of other applications (like agricultural and industrial vehicles) that would require much simpler control mechanism.

Many agricultural vehicles are computer-controlled when in the field. They cost so much, however, that the additional cost of the operator's time is fairly insignificant.

> I really feel autonomous cars are a prime candidate for Next Big Thing over the coming 10-25 years.

Very likely, and overall a good thing™. I'm waiting to hear when the first self-driving car runs over a motorcyclist or bicyclist, then says, "Hey! I didn't see that little thing!"

In a sad commentary on modern times, auto fatalities continue to decline, while motorcycle and bicycle fatalities continue to climb.

Who is driving 100mph?

The one thing that sticks with me in all of these automated driving demonstrations is the lack of operation in adverse environmental conditions such Rain, Fog, Snow/Ice. I would expect that any one of those conditions would post extraordinary challenges to an automated driving system.

Interesting, I have exactly the opposite view. The limited visibility that such conditions enforce, making them hard for humans to handle, have much lower effect on a system equipped with "vision" on multiple wave-lengths, for some of which the rain and fog are as transparent as air. As for the other effects (say, less tire friction) I don't see why a computer would do worse. I'm pretty sure the breaking distances are not hard-coded, but rather a very sophisticated feedback loop.

From what I understand, Google's self driving car is heavily based around the lidar sensor. The lidar sensor is used to image immediate surroundings in 3D in real time. Rain, snow and fog interfere with the laser pulses sent out by the lidar sensor, which basically renders the self driving system blind.

You are correct. The system degrades in functionality in dust, heavy fog, heavy rain, and snow. It isn't blind, per se, as radar still functions. Snow is the worst though - both visually obscuring and radar absorptive.

We're very far from this being the case now, though. Humans are very good at filtering multiple inputs and coming up with a reasonable understanding of their surroundings when sensory input is limited or different form normal. Computers are still not. Yes, maybe different sensors on the car will help, but real situational awareness and adaptability is what will be hardest to program. Not something that will never happen, I'm sure, but also not as soon as some hope.

Yes, this one factor will be the key to the success of this kind of technology. As much as I love Google spearheading this movement, other car companies have made good headway into catching up in a way that is more robust to rain/fog etc and also doesn't need pre-mapped routes.

Mercedes: https://www.youtube.com/watch?v=LHqB47F12vI Note the lack of GIANT lidars :). Though I'm not entirely sure how much of it is really autonomous.

I agree! This is why I'm more excited about Volvo's technology than Google's: I'm sure people in Sweden have snow at the top of mind when developing this technology, rather than sunny Californian roads.

I believe the autonomous cars simply can't operate in adverse conditions and have mechanisms in place to detect when the sensors are failing. This is why they still require an operator in the driver's seat. It's also why the car they showed without the steering wheel is a bit ridiculous.

I still think autonomous cars are far far away from being used on a daily basis, especially in other parts of the world - right now you need good weather, precise maps, well maintained roads and the cars are taken care of by a team of engineers. I just can't imagine such car, for example, in Italy - you would need to replace all cars at once, because people drive there like insane and an autonomous car wouldn't even move.

What's interesting is that the reported test is from 2012. (Which they do say at the top, but didn't really register until the end.) I'm a bit surprised it was so competent back then.

After getting my first traffic-camera ticket yesterday, I'm 100% behind self-driving cars.

I hate situations in which the traffic light turn red just before I cross the street.

Most cases, I made the correct choice, but sooner or later I am going to screw up.

But a self-driving car shouldn't get any hangup like that and the AI should improve each time it gets into an accident just like airline safety improve after each accident.

> I hate situations in which the traffic light turn red just before I cross the street.

In the UK there are amber lights for this purpose. The light changes from green to amber, before turning red. Both amber and red mean 'stop', but you're not going to get a ticket for going through an amber light.

"You MUST stop behind the white ‘Stop’ line across your side of the road unless the light is green. If the amber light appears you may go on only if you have already crossed the stop line or are so close to it that to stop might cause a collision."

(From https://www.gov.uk/using-the-road-159-to-203/road-junctions-...)

The US has the same thing the problem is it's not always followed. Many drivers don't view the amber/yellow light as "I should stop now" but rather view it as "I better speed up so I don't get stuck at the red light". There's also been much accusation (though I have no links handy to provide as proof) of townships implementing red-light cameras and then changing, without warning, the duration of the amber light to be shorter. Keeping in mind that the law states that you cannot be inside the intersection when the light is red, if you entered it while the light was green or amber is irrelevant [Edit - It's been correctly pointed out below that this law may vary from state to state].

That's not the law in California. It is perfectly legal to be within an intersection when the light turns red. It's only illegal to enter an intersection while the light is red.

Source 1: California Driver Handbook

http://apps.dmv.ca.gov/pubs/hdbk/traff_lgts_sgns.htm

- - -

Solid Yellow – A yellow signal light means "CAUTION." The red signal is about to appear. When you see the yellow light, stop if you can do so safely. If you cannot stop safely, cross the intersection cautiously.

- - -

The "stop if you can do so safely" wording is safety advice, and sound advice too. But it's not a statement of law.

Source 2: California Vehicle Code

http://www.leginfo.ca.gov/cgi-bin/displaycode?section=veh&gr...

- - -

21452. (a) A driver facing a steady circular yellow or yellow arrow signal is, by that signal, warned that the related green movement is ending or that a red indication will be shown immediately thereafter.

21453. (a) A driver facing a steady circular red signal alone shall stop at a marked limit line, but if none, before entering the crosswalk on the near side of the intersection or, if none, then before entering the intersection, and shall remain stopped until an indication to proceed is shown, except as provided in subdivision (b).

- - -

Note that the yellow light is described only as a warning, with no mention of it being illegal to enter the intersection while the light is yellow. And there is no mention of it being illegal to be inside the intersection when the light turns red, only that you can't enter on a red light.

I can't speak for other states, but I suspect that the laws in many of them are similar.

Thank you for the correction. I'm in NJ and remember my instructor going out of his way to mention that it was illegal to be in the intersection when red. However upon investigation I can't actually find anything to back this up now. It would appear I may have been operating under false information all these years.

I guess that's not an unreasonable thing for a driving instructor to say, since even if it's not quite true, it does no harm to believe it and may well help someone err on the side of safety.

I had a similar misunderstanding myself until recently. I'd read the wording in the driver handbook and thought that the bit about "if you can stop safely" was the actual law rather than safety advice. It was much later that I got curious and looked into it a little further.

It's not quite as clear as that, the California Driver's Handbook states:

"Do not enter the intersection if you cannot get completely across before the light turns red. If you block the intersection, you can be cited." [0]

You might think this is only referencing situations where you are stopped in the intersection (i.e. blocking it) but there is no reference to speed here.

There's a question on the test: [1]

Q: It is against the law to enter an intersection when:

A 1: You can't get all the way across before the light turns red

A 2: The light is flashing yellow and you didn't stop first

A 3: The light is yellow and you can stop safely

The correct answer is 1.

So, when viewed by a normal person, you may assume that this only applies if you're forced to stop in the intersection (and therefore block it), but neither the driver's handbook, nor the test itself (sorry my link is just an unofficial practice test, but it's on the real thing too) make any reference to speed. Therefore I think it's absolutely reasonable for your "instructor going out of his way to mention that it was illegal to be in the intersection when red" [GP]

I haven't looked up the actual law to see if the wording makes it OK to be in an intersection with the light red, providing you entered legally (yellow/green) and have a clear exit path - but I'm guessing it does not.

[0] http://apps.dmv.ca.gov/pubs/hdbk/traff_lgts_sgns.htm

[1] http://www.elitedrivertrainingservices.com/tests/AdultPracti... (Q10)

Fortunately you can't be cited for violating a private company's example test question! :-)

If you block an intersection because there isn't room on the far side, you'd be cited under section 22526 of Chapter 9, Stopping, Standing, and Parking:

http://leginfo.legislature.ca.gov/faces/codes_displaySection....

- - -

22526 (a): Notwithstanding any official traffic control signal indication to proceed, a driver of a vehicle shall not enter an intersection or marked crosswalk unless there is sufficient space on the other side of the intersection or marked crosswalk to accommodate the vehicle driven without obstructing the through passage of vehicles from either side.

- - -

This section is known as the Anti-Gridlock Act of 1987. The reason this had to be enacted was that it was never illegal to be in an intersection when the light turned red. If that were illegal, then officers would already have a section number to cite you under. They didn't, so this law gave them one.

Besides that, the only other relevant sections I could find are the ones I quoted in my previous comment. Here's another link in a newer format:

http://leginfo.legislature.ca.gov/faces/codes_displayText.xh....

The rest of the Vehicle Code can be found through a link on this page:

http://leginfo.legislature.ca.gov/faces/codes.xhtml

[directionless rant that I'm going to post any way]

I've skipped red lights, not just to make way for ambulances and police cars (but yes that) but also when the amber->red transition is very short, you're doing the speed limit and when you check your mirror prior to putting on the brakes the driver behind is seemingly trying to read your GPS or check themselves in your rear-view mirror [ie is too close]. Slamming on the brakes then is dangerous; that's an offence too, damned if you do ...

The "must" in "You MUST stop" is an interesting one. The Highway Code is basically [correct me if I'm wrong] a summary of a code that if applied correctly should keep you from being in breach of the Road Traffic Act/ other similar Acts. So it says you must stop for a red light or you will breach, for example here, RTA Section 36 (http://www.legislation.gov.uk/ukpga/1988/52?view=plain).

Now reading RTA Chapter 52 Section 36 I find it tells me about traffic _signs_. So, wither traffic lights, well the interpretations at S.192(1) are useful and refer me to Road Traffic Regulations Act (http://www.legislation.gov.uk/ukpga/1984/27/section/64) but still not finding anything that tells me about traffic lights.

The details in RTA 36 concern proper placement and use of a sign, but this doesn't really seem to cover the complete situation - clearly a policeman on duty could direct you to cross a line at a red-light, or indeed the Highway Code says that when lights fail (and you know that how!?) then you can ignore them.

So, traffic lights are a special case of traffic signs that don't really get well accommodated it seems. I concede I gave up when I hit the fifth "reference this other section for what a 'traffic sign' is" so there could be detail elsewhere.

/rant

PS: Ah, http://www.legislation.gov.uk/uksi/2002/3113/regulation/10/m... = TSRGD regs 10 & 36, does the bit about traffic lights. Apparently only a red light is a sign under S.36 above.

> * cases, I made the correct choice, but sooner or later I am going to screw up.*

In Australia, when you screw it up, you get a ticket for the red light and a ticket for speeding (if you were).

It's not uncommon to lose your license for that "one" incident.

The punishment of losing a license is one reason I think self-driving cars should be allowed to operate without one. It would be locked into automatic mode or allow a restricted (low speed only) manual mode for emergencies.

This would also help the elderly or disabled who would not be able to pass a driving test.

Reply on news.ycombinator.com