Hacker News Reader

Comment by manicdee | Hacker News Reader

"They drive funny" claims human, regarding high level of reporting of accidents where computer drivers are required to report encounters but humans are not.

If humans were required to report every bump and scrape, what would we find out? I suspect we would find that humans are just poor drivers overall, and "they don't drive like I do" is the number one cause of minor incidents worldwide.

My main submission to this argument is the ubiquitous rear-ending in the turning lane where the first car's driver starts to roll and then stops, while the second driver is turned to watch approaching traffic and thus oblivious to the behaviour of the car in front.

View on HN

> If humans were required to report every bump and scrape, what would we find out?

This seems like an awfully good point. One of the incidents described was a bicycle rear-ending a car - reporting from that incident reveals a scraped knee, a scratched fender, and an intact bike. How many times a day do equally minor accidents between people get ignored?

Maybe self-driving cars really do get rear-ended more, but I haven't seen anyone try to account for reporting gaps like that.

http://www.thedrive.com/tech/11407/a-self-driving-chevy-bolt...

The issue isn't that the cars are worse drivers than humans. They're obviously going to be much better statistically. The article is just making the point that they're going to have to be programmed to break the rules just like humans do i.e. with speeding, stopping too often etc. If they are not programmed to break the rules, they will frustrate people and even cause accidents as humans try to interact with them. There's not always a good way to know that the car in front of you is being driven by a computer.

This will be an opportunity to fix the traffic rules to be more realistic. Often, the speed limits are too low in the US, and "everybody" goes faster. This leads to the problem of selective enforcement, and people not trusting speed limits when they are really neccessary.

In contrast, I've found most speed limits in Germany to be reasonable absolute maximum limits - except on one or two places on streets I know very well. And I'm not talking about the Autobahn - if I see 70 km/h in Germany (45-ish mph), its probably dangerous to go faster. If I see 45 mph in the States, it seems to mean "everybody go 55 mph".

Counterpoint: as a German, I know that when a sign says 50kmh, everyone goes 60, and when it says 70, everyone goes 80, etc. If you go as fast as the sign says, you will most likely cause traffic.

> If you go as fast as the sign says, you will most likely cause traffic.

Counterpoint: creating traffic by going at the speed limit requires that other people on the road are going over. Both parties are to blame, but only one party isn't increasing their energy at a exponential rate (1/2mv^2).

> Often, the speed limits are too low in the US, and "everybody" goes faster.

Excessive speed is the primary cause of a third of all fatal accidents and a contributing factor in at least half. Instead of training the self-driving cars to be more human, we ought to train humans to act more like the computers! Speeding is one of the stupidest norms that exists in our society. We should definitely not try to encourage it.

No, difference in speed causes those accidents:

https://en.wikipedia.org/wiki/Solomon_curve

> Speeding is one of the stupidest norms that exists in our society.

Politicians arbitrarily setting the speed limits is the root cause.

The EU has many countries which have higher freeway limits, yet lower accidents per mile driven. They also have more stringent licensing.

Yeah, that's not a great study (though it was admittedly pretty groundbreaking,) and it doesn't actually prove that there is no relation between speed and accidents. Multiple studies since the 60's have shown different results.

>Politicians arbitrarily setting the speed limits is the root cause.

Of speeding, or accidents?

> Yeah, that's not a great study (though it was admittedly pretty groundbreaking,) and it doesn't actually prove that there is no relation between speed and accidents. Multiple studies since the 60's have shown different results.

The link sourced multiple studies. The originals were dubious, subsequent ones not so much.

But the issue the Solomon curve and other similar studies points out is that humans (and largely Americans) will ignore the rules if they feel they can safely go faster.

And sure, you can't ignore overall average speed. Crashes on roads where the average speed is 100 mph are going to be much more prone to fatalities than 55 mph roads. So, there are two rough equations: Lower average speeds mean less severe crashes; speed limits set to the 85th percentile mean fewer crashes. As the speed limit is raised (from 0), there will be at least one solution that satisfies both of those equations and results in the fewest fatal crashes (if that's your goal).

In certain situations the lower speed limits can make things less safe. For example, a straight, well maintained toll road with wide lanes, clear vision ahead, and say wildlife barriers that was designed for 75mph max speeds is constructed in a state with statewide 55mph speed limits. Construction uses the latest traffic easing and safety techniques, since you don't want to construct all roads to enforce speed limits that may change in the future.

So now most truckers and cautious drivers go 55mph, while people with less inhibitions are weaving in and out of them.

Are you talking about the wikipedia link? All the subsequent studies save one that it sourced disproved the first's premise or drew attention to serious questions as to its validity. Did you read them?

Yes, but the speeding norm is partly caused by speed limits being uniformly too low in some circumstances leading to speeding being tolerated to a point and everyone making their own judgements about how fast is too fast. It would be better to have higher limits that are strictly enforced rather than lower limits that no one follows.

Did you bother reading what he said? Speed limits are lower than what is safe on many roads. Limits should be increased to the safe maximum and then the maximum should be more strictly enforced. Minimums should be enforced as well, since driving significantly below the maximum is also unsafe.

It doesn't mean increasing limits to something that is unsafe.

1. Do you want to be the local politician who says "hey, let's raise this to 55 MPH" and then on a bad weather day or just out of sheer bad luck soon after there's a major pile up and then said politician gets roasted?

1. The traffic lights need to be coordinated and that means presuming a certain speed. Now the problem is a lot of traffic lights are close enough that changing the programming of just one should ripple through the entire system and this recently (especially on a regulation making time scale) was a problem consuming extraordinary amounts of computing power.

Speed limits aren't just about safety. Lower speeds mean less noise, which is desirable to local residents in residential areas. They're the ones who local politicians are accountable to.

Pollution too. Speed limits are sometimes lowered when it is a problem. Temporary, during peaks, or permanently.

The other, more cynical reason for lowering speed limits is that in many countries now, speed traps are a significant source of income for the state. Making the speed limit significantly lower than what the standard 85% rule says is a good way to extract money from unsuspecting drivers.

Decreasing speed limits is something that local politicians can use to demonstrate to their constituents that they care about them, but very rarely has net-positive effect on noise and polution.

but this is exactly the sort of political cover the politicians need. nobody wants to be the politician that says "lets raise the speed limit here to 55MPH", but if everybody drives 55 and all of a sudden there's a bunch of self-driving cars going 50 and people are getting annoyed at the road getting blocked up, changing traffic rules to conform to reality becomes a lot more possible.

Your interpretation of American speed limits is quite correct. Driving the speed limit on a major highway in any but the rightmost lane is dangerous due to the speed other drivers expect you to be going.

On the flip side (almost literally), I once drove around Ireland for a week. Of course my instinct was to drive slightly over the speed limit, but I quickly found 100 kph on 1.5-lane roads winding around sheep pastures with no visibility nearly impossible to safely match let alone exceed…

>The issue isn't that the cars are worse drivers than humans. They're obviously going to be much better statistically.

This isn't obvious at all.

Sure it is!

There’s a way to know they’re learning to drive or just recently got their license, why wouldn’t there be a way to know they’re driven by software?

Right, just give them a C plate.

That, or... if enough AI drivers are out there then maybe humans will just learn to live with them.

While crashing into an AI driver I may be able to blame the AI, it's still a huge pain and possible loss for me as a human.

If the reality is that AI is not leaving and they have peccadilloes, then the human drivers and legal structure will just need to adapt (something we aren't bad at).

Computers have been training humans for the past couple decades, and I see no reason for that to stop just because the computer's form factor is a car.

They should follow the law. Granting an exception is not the way the law should work, because such exceptions are arbitrary and without any sense of justice or ethics. The best way to fix a bad law is to consistently enforce it.

If two actors, one of which is following the law, has a higher risk than two actors both breaking the law, the lawbreaker is still at fault for the increased risk.

Speed limits are not just for traffic safety, but also to reduce noise pollution.

If you take a system where everybody does similar and add something that does not, this is going to cause problems. This is going to cause problems even if the dissimilar behavior is the proper legal behavior.

Erratic behavior causes accidents, be it driving too slow, too fast, weaving, impaired, etc...

Now, if all the cars drove the same way, we'd be fine. We'd have very few accidents. However, we don't. So, we bump into stuff.

Erratic behavior, due to following traffic laws?

From the point of view of drivers who're used to everybody bending the rules to a certain level, strictly following the rules is at the very least unexpected; 'erratic' is perhaps an overly strong word but I think the general point is sound.

If it helps, you can call it things like non-standard.

We don't actually tend to follow all the rules of the road. Like all changes in traffic, we will adapt. It takes time. Something like a rotary can take 18 months before traffic fully adapts. I can't speculate as to how long this will take.

"They aren't driving like I would" is a major red flag for me, causing me to either drive/bike/walk more carefully, get well away if possible, and keep them in view if necessary.

Sounds like success to me.

I've been in 3 minor car accidents, ALL 3 have come where someone read-ended me in this way. I am constantly conscious of this and am always keeping an eye on the car behind me when I'm in this situation.

This is why Automatic Emergency Braking (AEB) is the most important safety feature since the advent of the seatbelt. It's going to save many lives.

Yes, it's relatively simple things like this that can actually improve car safety right now, no self-driving capabilities necessary. Automatic braking, lane departure warning, dead-spot detection, driver attention detection, missed signal warnings, etc.

I don't understand why we are not investing all of the efforts and resources that are now spent to develop full self-driving capibilities, into making existing viable safety technology cheaper, and eventually mandatory. I'm 100% convinced it would have a much bigger impact on traffic safety, in a much shorter timeframe, to the point we don't even need self-driving cars anymore. At least not for safety purposes.

Come to think of it, maybe the whole self-driving car hype is not actually because companies want to improve car safety, but because they see new business models based around them, and we, the public, need to be convinced self-driving cars will be safer than alternative technologies or modes of transportation.

Yes, AEB and Adaptive Cruise are the two most important features I looked for in a car. I ended up leasing an EV without them (waiting for my Model3), but my next vehicle will require these (for example, almost all Toyota models have similar features - TSS-P warns for frontal crash but doesn't brake immediately).

Since beginning of this year I almost exclusively drive cars with some kind of front facing radar for AEB/ACC. On one hand there were situations where imminent collision warnings or even AEB were useful, on the other the amount of false positives for collision warning is annoying and in some cases actually dangerously distracting.

And for some reason the minimal front distance for WV's ACC is almost an order of magnitude larger than safe braking distance, so other drivers see no issue with changing lanes into that large free space in front of you.

So in all, this technology obviously requires some work both in terms of reliability and in terms of UX.

It's certainly because of the potential market. Nobody is even claiming to be researching self-driving cars for the safety, it's always because of the potential.

Two thoughts:

1. Why don't people pay attention to break lights? It's hard for me to relate to people who are always stopping last minute.

2. When someone is driving aggressively or erratically behind me, I break-check them as often as a I feel I need. Won't solve all rear-end issues, but it improves your chances of avoiding accidents.

Pretty sure it'll increase your chances of being IN an accident.

One of the most common accident modes is right-turn-on-red. The first car will stop (or not), creep forward, etc. If they see an opening -- or think they seen an opening -- they'll let off the brakes and begin to roll under idle power or even accelerate slightly. Of course, this is the cue that they're making a right on red, so the 2nd car in line will also accelerate and at the same time look to the left for oncoming traffic. And BAM, you rearend the car in front of you because you only thought it was pulling out. It's the perfect storm of misreading the situation, and the situation demanding that you look somewhere other than where you're going.

Come to think of it, I've been hit from behind this way twice, and done it to someone else once.

I'm sure self-driving behavior is a factor here, but I also suspect that breaking and passing at intersections is just a common cause of zero-damage accidents. After all, one of the collisions in the story was a bike rear-ending a car - how many drivers would report that if there was no damage?

if someone is going to rear-end you, there's not much you can do (even if you can see it coming), other than brace yourself

I've heard (citation needed) that if you are going to be in an accident the best step to take is to relax.

Anecdotally, it depends on how far away you see it coming. I'm constantly scanning front, rear mirrors, and side mirrors and if someone is driving erratically I take steps to get out of their way (in a safe and predictable manner of course!) I have no evidence this keeps me safer, but this and other "traffic calming" behavior I engage in makes my commute less stressful. It has the nice side-effect of continuous practice that reinforces good driving techniques.

"if someone is going to rear-end you, there's not much you can do (even if you can see it coming), other than brace yourself"

.. and by "brace yourself", you would touch your head to your headrest so that your neck has nowhere to move/flex with the impact. You want all of your body to be completely rearward in your seat.

Also, while this may be morally negative, if there is another car in front of you, you could creep forward and lightly touch their car with your own, making yourself the middle ball in the Newtons Cradle[1] that your three cars now describe.

[1] https://en.wikipedia.org/wiki/Newton%27s_cradle

Unless you can guarantee contact, is moving towards another car a good idea? Intuitively (which of course could be _completely_ wrong), it doesn't seem like being forced to decelerate sooner would be useful.

Hrmmm. I've avoided two by knowing the stop was coming and watching the rear view -> move to shoulder.

Yes, but that is avoided by years of experience and pattern learning, something AI should be able to figure out quickly.

Reply on news.ycombinator.com