Autonomous Cars

[AJI]

...We humans are pretty poor at doing these calculations preferring narrative stories to numbers.

This^

Can we please get back on track?

[Phil]

More than 3000 people die on the roads... DAILY!
https://www.asirt.org/safe-travel/road-safety-facts/

What did you expect, with an estimated 1 billion cars on the road, every day?

What does that make 3000 deaths, roughly, 0.0003%? These sort of scale arguments don't really work or prove anything, not least because there's an insignificant amount of AVs currently (a single sand grain on a beach basically). It would make more sense to stick to a general country like the US and analyze the numbers there. My bet it is quite a bit lower than 0.0003%, not least because there are actual laws. There's probably a statistic somewhere that shows the correlation between deaths/accidents and laws, not to mention the severity of punishments if laws are broken (e.g. speeding). Did you know there are countries that allow you to drive without basically any training at all?

Reaction times are reduced to milisecods instead of tenths (or seconds if the driver is distracted), increasing the oportunities to avoid the accident
Constant 360º monitoring
Vehicle and pedestrian detection even at a dark night without any light, with fog, etc

General reaction times are more around a second at least for humans, not tenths. More like 2 seconds if something unexpected happens. That's without distractions. You need to see, process, react and the movement of your foot to the brake pedal takes time too. Of course, an observant driver might be better prepared for certain eventualities. The warning sign of a school nearby might lead to a reduction of speed and the readiness to brake to lower the reaction time perhaps to a second.

While you (and others) grand solution is to take away control of all humans and replace it with some software, why not simply add assists to the car itself? We already have this in the form of brake assist for example. The car constantly monitors the distance to the car in front; if you drive too close to it, it will signal a warning to inform the driver that he is too close and that in case of an emergency braking, there is not enough time to avoid a collision. Most humans also fail to apply emergency braking (the force required to do so is immense), thus why most/many modern cars have brake assist. You have ABS to optimize grip and braking distances. You could extend this to the point that a driver has better technologically assisted tools for more eventualities. Yes to sensors for better situational awareness, but no to full autonomy. The best of both worlds.

You might be afraid of drunk drivers - simply add (by law and regulation) an alcohol tester to the person driving the car. In my country, the laws have been tightened in regards to speeding and alcohol with potential jail time. It works.

[Zynerji]

I´ve just deleted a very long reply because these sort of debate searching for a fifth cat leg bore me to death.

I think any vehicle moving within the speed limits, constant 360º monitoring with no dead angles, and almost instant reaction time will never be on a situation where it has to choose who will die, both if the driver is a human or a computer. Computers can guarantee this, humans can´t. To me it is this simple.

That hypotetic scenario is possible, but I think we will only see it in the news as some odd accident while deaths on roads and streets decrease dramatically because now drivers are, all of them, focused on the road and driving properly... because they´re computers

Disclaimer: I´ll hate that moment, I love driving more than any of you will believe, both cars and bikes, but I think that´s what future will bring us

Why would Germany already be setting rules about it if it weren't a realistic problem to overcome?

5th cat leg...

I guess as long as they are not 737 flight computers, they should be perfect...

Because as I´ve stated that situation might be possible, so the programation must be taken into account. But that does not mean the situation will be usual, I´m sure 99% of people will never live that, period, but for any newspaper that debate is gold and will provide some thousands clicks

Do you really think a car moving within the speed limits, with 360º monitoring and instant time reaction will be on a situation where it must choose who will die Zynerji? Be serious considering the question please

Its not about what happens in a controlled situation to me, it's the weighting of the variables in an uncontrolled situation that bothers me.

I'm only 40, but I have witnessed enough utterly crazy things that drivers do, and the transition from self driving to auto driving is going to include much of it. I just hope that the Lawful are protected specifically over any other concern, as that is the only "fair" variable to judge.

And NO, I do not believe this will be a thing at 100% AV control, as all variables will be controlled, and 99.9% of accidents will be prevented.

[Phil]

Zynerji,

You raise some good point, but to be honest, I think many of the things you are arguing about (making the decisions of rescuing ones own life compared to others) come down to instinct. In such a situation, there's no time for logical, rational thought. If kids, no matter who is right or wrong, jump in front of your moving vehicle, you react without thinking. And in such collision, no matter the outcome of who lives and who doesn't - there will always only be losers, arguably for those that survive more so then they who don't, because they will have to live with the consequence.

On the other hand, the point that a piece of software will take over this decision based on any rational (e.g. who is right/wrong, who's lives are deemed worth more) will not happen. Not for a very very long time. It's arguable it's even possible to create consciousness (A.I.) - until then, it's all just stupid software with a lot of IF-clauses in lots of code. Situation Y, do Z etc. There won't be software that's so complex that it will take into account these complex questions. You can be happy if AV ever reach a point that they can drive in a complex environment and can deal with all kinds of unpredictability while remaining perfectly safe.

[Andres125sx]

More than 3000 people die on the roads... DAILY!
https://www.asirt.org/safe-travel/road-safety-facts/

What did you expect, with an estimated 1 billion cars on the road, every day?

What does that make 3000 deaths, roughly, 0.0003%? These sort of scale arguments don't really work or prove anything

So we must stop investigating for new active and passive safety measures? You think that´s an aceptable rate of deaths a year for a transport?

Reaction times are reduced to milisecods instead of tenths (or seconds if the driver is distracted), increasing the oportunities to avoid the accident
Constant 360º monitoring
Vehicle and pedestrian detection even at a dark night without any light, with fog, etc

General reaction times are more around a second at least for humans, not tenths. More like 2 seconds if something unexpected happens. That's without distractions. You need to see, process, react and the movement of your foot to the brake pedal takes time too. Of course, an observant driver might be better prepared for certain eventualities. The warning sign of a school nearby might lead to a reduction of speed and the readiness to brake to lower the reaction time perhaps to a second.

I know, I was just trying to avoid someone arguing "no, humans can react much faster than a second". But yes, I know difference in reaction times between a computer and a human are actually much higher than I stated.

So I guess you agree only because of this, AVs will reduce crashes drastically when compared to humans. On a dangerous situation like a kid jumping on the road one second may easily be the difference between an emergency stop with no consequences, or having to choose between running over the kid, crashing with a car in front or a tree, or driving off of a cliff

While you (and others) grand solution is to take away control of all humans and replace it with some software...

Not true Phil. I´m not proposing anything, I´m just saying what I think future will be, as all of us are doing in this AV thread, as AVs do not exist yet. I mean an L5.

why not simply add assists to the car itself? We already have this in the form of brake assist for example. The car constantly monitors the distance to the car in front; if you drive too close to it, it will signal a warning to inform the driver that he is too close and that in case of an emergency braking, there is not enough time to avoid a collision. Most humans also fail to apply emergency braking (the force required to do so is immense), thus why most/many modern cars have brake assist. You have ABS to optimize grip and braking distances. You could extend this to the point that a driver has better technologically assisted tools for more eventualities. Yes to sensors for better situational awareness, but no to full autonomy. The best of both worlds.

Agree, that would be enough IMO, but you know AVs will be a reality sooner or later, right? Probably at first just as a luxury product for those who don´t like driving and prefer investing that time on some other thing while the car drives itself. But then, as always, more and more people will bit, and at some point I think they will be mandatory, not because I think they must be, or because I think that will solve anything, but because I think that´s what will happen.

But you´ve just opened a new window to me I didn´t consider before. This will be progresive, and meanwhile more and more assistancess will be included in our daily cars. Hopefully those will reduce deaths enough to avoid some moron on a political charge doing what they always do, and we will always be allowed to drive manually.

[Zynerji]

Zynerji,

You raise some good point, but to be honest, I think many of the things you are arguing about (making the decisions of rescuing ones own life compared to others) come down to instinct. In such a situation, there's no time for logical, rational thought. If kids, no matter who is right or wrong, jump in front of your moving vehicle, you react without thinking. And in such collision, no matter the outcome of who lives and who doesn't - there will always only be losers, arguably for those that survive more so then they who don't, because they will have to live with the consequence.

On the other hand, the point that a piece of software will take over this decision based on any rational (e.g. who is right/wrong, who's lives are deemed worth more) will not happen. Not for a very very long time. It's arguable it's even possible to create consciousness (A.I.) - until then, it's all just stupid software with a lot of IF-clauses in lots of code. Situation Y, do Z etc. There won't be software that's so complex that it will take into account these complex questions. You can be happy if AV ever reach a point that they can drive in a complex environment and can deal with all kinds of unpredictability while remaining perfectly safe.

https://towardsdatascience.com/implemen ... 6780bc70f4

I think you might be a bit behind, because this is really happening. Tensorflow+Tensor cores+250k sensors= viable real time behavior control and prediction.

Honestly, they should pay the insurance companies to put dash-cams in their safest drivers cars (huge discount) to record the habits of the best available human drivers to train these models. That would guarantee that it is operating off of human values and intuition, then let the model determine the finer granuality of the behaviors.

[strad]

[Andres125sx]

Why would Germany already be setting rules about it if it weren't a realistic problem to overcome?

5th cat leg...

I guess as long as they are not 737 flight computers, they should be perfect...

Because as I´ve stated that situation might be possible, so the programation must be taken into account. But that does not mean the situation will be usual, I´m sure 99% of people will never live that, period, but for any newspaper that debate is gold and will provide some thousands clicks

Do you really think a car moving within the speed limits, with 360º monitoring and instant time reaction will be on a situation where it must choose who will die Zynerji? Be serious considering the question please

Its not about what happens in a controlled situation to me, it's the weighting of the variables in an uncontrolled situation that bothers me.

And I agree that´s a concern. Only that I think uncontrolled situations will be marginal, and that will be a huge improvement, even if in those situations reactions are not perfect. Are they today? Obviously no, many drivers today simply make no decision at all when they face an unexpected danger and freeze. Others will react properly, and others will react but will be unable to control the car and will crash with some other car, pedestrian or whatever.

I think we´re approaching this differently. You seem to take it as will AV be perfect in all situations?, while I take it as will AV be a step forward?. They will be far from perfect obviously, and will cause some situations to consider, but I think the benefit may be so huge those concerns about very odd situations will be minor compared


See it this way, I do prefer if an uncontrolled car in front of me is driven by a computer programmed with some parameters we humans made an agreement, than a random human at the wheel

And NO, I do not believe this will be a thing at 100% AV control, as all variables will be controlled, and 99.9% of accidents will be prevented.

I don´t get what you mean here. My english is poor sorry

[hollus]

3000 a day. Can that number really be correct?
It adds up to roughly 1.000.000 a year, which, over a period of 70 years (a life expectancy) builds up to 70M. That’s almost 1% of the world’s population!
Is it really that high?

[Phil]

https://towardsdatascience.com/implemen ... 6780bc70f4

I think you might be a bit behind, because this is really happening. Tensorflow+Tensor cores+250k sensors= viable real time behavior control and prediction.

Honestly, they should pay the insurance companies to put dash-cams in their safest drivers cars (huge discount) to record the habits of the best available human drivers to train these models. That would guarantee that it is operating off of human values and intuition, then let the model determine the finer granuality of the behaviors.

Of course my post was grossly simplified. I am well aware of what technology is currently capable of and yes, it's impressive. But in the end, it's still all software with zero intelligence behind it. I've already stated in this topic that I think a centralized software that controls the entire traffic flow of all cars would work quite good. This would probably also cut down the need for too many sensors on the cars too, given most things moving will be controlled anyway. This is assuming there are no more driven cars. A solution where the software is inside the car itself and can communicate with other cars around (think TCAS but for cars) might do the trick too. Once you have humans sharing the road, predictability becomes a whole new ball game. Anyone who drives on the road could write books about erratic behavior.

A smart software would take the worst case scenario into account at all times which would render it pretty incapable of driving at any sufficient speed. The same applies to driving yourself - if you had to assume there's a kid waiting to jump out and into your direct path at any moment, you'd probably quit driving for good.

Developing software to counter every single unpredictability (humans) on densely populated roads is extremely difficult and will require jumps an order of magnitude higher. Again; not saying it isn't impressive what currently is being done. I'm simply saying what will be required is immense.

One interesting point though; How would a software realize that what it mistakenly recognized as being a motorbike is in fact a young kid on an bicycle (an e-bike)? Or a tree that is in fact a person wearing some funky clothes? A newspaper flying across the road and into your path by the wind that the car mistakenly recognizes as a foreign unknown object, an animal whatever? Or at the next instance, that same newspaper but in fact is a small kid wearing a grey raincoat running onto the road?

To Andres and others; There's no dispute a machine can and will react quicker than a human. That's not what I am disputing. I am dreading the car not identifying what it should and causing an avoidable collision and possibly death while a human is sitting there that could have easily prevented it knowing instantly what it was with time to spare.

[Zynerji]

LORA RFID tags on everything!

[henry]

3000 a day. Can that number really be correct?
It adds up to roughly 1.000.000 a year, which, over a period of 70 years (a life expectancy) builds up to 70M. That’s almost 1% of the world’s population!
Is it really that high?

No it’s higher. 1.25 million in 2013 and rising. https://www.who.int/gho/road_safety/mortality/en/

[henry]

3000 a day. Can that number really be correct?
It adds up to roughly 1.000.000 a year, which, over a period of 70 years (a life expectancy) builds up to 70M. That’s almost 1% of the world’s population!
Is it really that high?

The 1% number is about right for the US, which is below average for the world. Here in the UK it’s about a quarter of that, and in Africa more than twice. To pick some numbers at random.

Another number of concern is the cost of RTAs. Here in the UK the cost is put at 2% of GDP.

[Andres125sx]

3000 a day. Can that number really be correct?
It adds up to roughly 1.000.000 a year, which, over a period of 70 years (a life expectancy) builds up to 70M. That’s almost 1% of the world’s population!
Is it really that high?

Yes they´re correct, but you know, this sort of scale arguments don´t prove anything for some people

If a terrorist attack causes 2000 deaths, on a specific date, that´s good to start a WAR. But if 3000 people die a year, every year, we should do nothing at the respect...


Numbers are dangerous, you can get used to them easily and assume they´re normal, even if they´re showing a terrifying statistic

[Zynerji]

3000 a day. Can that number really be correct?
It adds up to roughly 1.000.000 a year, which, over a period of 70 years (a life expectancy) builds up to 70M. That’s almost 1% of the world’s population!
Is it really that high?

Yes they´re correct, but you know, this sort of scale arguments don´t prove anything for some people

If a terrorist attack causes 2000 deaths, on a specific date, that´s good to start a WAR. But if 3000 people die a year, every year, we should do nothing at the respect...


Numbers are dangerous, you can get used to them easily and assume they´re normal, even if they´re showing a terrifying statistic

I hear we each swallow 8 spiders a year in our sleep as well.


Some statistics are beyond our control. I mean, every car has seatbelts, but roughly 60% of all road fatalities are unbuckled. No matter how safe/controlled things become, there will always be outlier statistics, and those that pay the ultimate price for their bad decisions.