When Will Electric Cars Become Mainstream?

[DougSpindler]

Maybe Tesla could add drunk and impaired to their list of driving options.

[b_force]

I find it always funny how these messages are getting blown up.
To put things in perspective, also look at similar accidents with human drivers and you will be amazed.

Remember: There's absolutely no limit to how bad human drivers can be.

Ref: https://www.google.com/search?q=bad+drivers&tbm=vid

At least robots don't get angry/drunk/bored/fall asleep.

Well I have seen enough official research, basically >96% of ALL accidents is because of human error.

So in all these news items about self driving cars I would really like to see a very decent analyses how humans would perform in the same conditions.
I highly doubt if that would be any better.

Humans would be perfored 99.99999% better.  And in the accidents where there as a fatality, the human drivers under the same cirucmstances would not have caused a fatality or even an injury.

Humans are good at somethings, machines are better at others.

And that is based on what research?
Like I said before, more than 96% of all accidents is because of human error.
Following that logic I find it VERY unlikely a human being would perform better.
Let alone so much better as you would suggest.

Unless laws of physics and statistics are biased somehow.

[DougSpindler]

I find it always funny how these messages are getting blown up.
To put things in perspective, also look at similar accidents with human drivers and you will be amazed.

Remember: There's absolutely no limit to how bad human drivers can be.

Ref: https://www.google.com/search?q=bad+drivers&tbm=vid

At least robots don't get angry/drunk/bored/fall asleep.

Well I have seen enough official research, basically >96% of ALL accidents is because of human error.

So in all these news items about self driving cars I would really like to see a very decent analyses how humans would perform in the same conditions.
I highly doubt if that would be any better.

Humans would be perfored 99.99999% better.  And in the accidents where there as a fatality, the human drivers under the same cirucmstances would not have caused a fatality or even an injury.

Humans are good at somethings, machines are better at others.

And that is based on what research?
Like I said before, more than 96% of all accidents is because of human error.
Following that logic I find it VERY unlikely a human being would perform better.
Let alone so much better as you would suggest.

Unless laws of physics and statistics are biased somehow.

Have you read any of the accident reports where a self-driving car resulted in a fatality?  In every case so far the human would have avidoed the accident.  The best documented accident of where the machine failed and the human driver was much better is the single car accident where the self-driving car ran into a tree killing the driver in California.  The driver had documented and reported the problem on a stretch of road where the car set-driving controlled car continually fails to stay on the road.

If you look at all of the other deaths involving a self-driving car you will find a human would have avoided them.  As for the non-fatal crashes, humans would have avoided over 99.9999% of those accidents too.

[b_force]

I find it always funny how these messages are getting blown up.
To put things in perspective, also look at similar accidents with human drivers and you will be amazed.

Remember: There's absolutely no limit to how bad human drivers can be.

Ref: https://www.google.com/search?q=bad+drivers&tbm=vid

At least robots don't get angry/drunk/bored/fall asleep.

Well I have seen enough official research, basically >96% of ALL accidents is because of human error.

So in all these news items about self driving cars I would really like to see a very decent analyses how humans would perform in the same conditions.
I highly doubt if that would be any better.

Humans would be perfored 99.99999% better.  And in the accidents where there as a fatality, the human drivers under the same cirucmstances would not have caused a fatality or even an injury.

Humans are good at somethings, machines are better at others.

And that is based on what research?
Like I said before, more than 96% of all accidents is because of human error.
Following that logic I find it VERY unlikely a human being would perform better.
Let alone so much better as you would suggest.

Unless laws of physics and statistics are biased somehow.

Have you read any of the accident reports where a self-driving car resulted in a fatality?  In every case so far the human would have avidoed the accident.  The best documented accident of where the machine failed and the human driver was much better is the single car accident where the self-driving car ran into a tree killing the driver in California.  The driver had documented and reported the problem on a stretch of road where the car set-driving controlled car continually fails to stay on the road.

If you look at all of the other deaths involving a self-driving car you will find a human would have avoided them.  As for the non-fatal crashes, humans would have avoided over 99.9999% of those accidents too.

You're using a loop argument, and yes there has been billions of similar accidents with humans.
The whole idea of the 96% number is that humans DO NOT always avoid accidents.
If you want to compare numbers you need to it fair and take a look at the whole picture.
Not just a few accidents that hit the news and think that's a representative number to prove that self-driving cars are bad or something.
That is at least not how thought me to do proper science and statistics.

Also it's not helping to bring that 96% number down.
So they have to look for alternatives right?

[coppice]

Have you read any of the accident reports where a self-driving car resulted in a fatality?  In every case so far the human would have avidoed the accident.  The best documented accident of where the machine failed and the human driver was much better is the single car accident where the self-driving car ran into a tree killing the driver in California.  The driver had documented and reported the problem on a stretch of road where the car set-driving controlled car continually fails to stay on the road.

If you look at all of the other deaths involving a self-driving car you will find a human would have avoided them.  As for the non-fatal crashes, humans would have avoided over 99.9999% of those accidents too.

In the early days of automating aircraft with software based flight control systems there were some horrendous screwups, which resulted in the total loss of aircraft. These were mostly fighter aircraft, so they incurred one death at a time. By the time large passenger aircraft were being automated the lessons from early mistakes in military aircraft had been learned, and things went pretty smoothly. Until recently the production and testing of the software for cars was also learning from these early mistakes. It feels like people have given up on this now. The recent issue with Tesla 3 braking distance is a good example. The public seems impressed that an OTA update released just a few days after the report of a serious safety issue has significantly improved performance. I'm not sure many people with a history in safety critical systems are so impressed. How well was this update tested for unintended consequences in such a short time?

[coppice]

You're using a loop argument, and yes there has been billions of similar accidents with humans.
The whole idea of the 96% number is that humans DO NOT always avoid accidents.
If you want to compare numbers you need to it fair and take a look at the whole picture.
Not just a few accidents that hit the news and think that's a representative number to prove that self-driving cars are bad or something.
That is at least not how thought me to do proper science and statistics.

Also it's not helping to bring that 96% number down.
So they have to look for alternatives right?

I'm not sure if you are missing the point, or trying to avoid it. Several recent incidents have shown serious holes in the systems in these autonomous cars. They aren't failing to avoid an accident, as a distracted human might. They are so poorly designed they aren't even TRYING to avoid some reasonably common crash scenarios.

[DougSpindler]

You're using a loop argument, and yes there has been billions of similar accidents with humans.
The whole idea of the 96% number is that humans DO NOT always avoid accidents.
If you want to compare numbers you need to it fair and take a look at the whole picture.
Not just a few accidents that hit the news and think that's a representative number to prove that self-driving cars are bad or something.
That is at least not how thought me to do proper science and statistics.

Also it's not helping to bring that 96% number down.
So they have to look for alternatives right?

I'm not sure if you are missing the point, or trying to avoid it. Several recent incidents have shown serious holes in the systems in these autonomous cars. They aren't failing to avoid an accident, as a distracted human might. They are so poorly designed they aren't even TRYING to avoid some reasonably common crash scenarios.

I agree he is missing the point.  How many years have we been working on automated-driving software?  Hasn't it been well over 50 years?  One would think after 50 years of develpement we are getting close to perfection.  Appears not...   As sophisticated as the software is I hope he can explain how the software mistook a tree for the road killing the driver.  And this wasn't the first time either.  There are at leat two accidents where the software drove the car right into a tree killing the driver.

And what about the times the software mistook a pakred police car and a parked truck for the road.

It appearss or poster hasn't learned from history, specificly the Therac-25 what can go wrong with computer programs. 
https://hackaday.com/2015/10/26/killed-by-a-machine-the-therac-25/

[Fungus]

Have you read any of the accident reports where a self-driving car resulted in a fatality?  In every case so far the human would have avidoed the accident.

You seem to be under the illusion that a zero-accident rate should be attained before deploying these.

That's never happened before with any new tech, why should it happen now? SO far they seem to be doing quite well.

  The best documented accident of where the machine failed and the human driver was much better is the single car accident where the self-driving car ran into a tree killing the driver in California.  The driver had documented and reported the problem on a stretch of road where the car set-driving controlled car continually fails to stay on the road.

I'm sure the software has been updated.

[coppice]

I'm sure the software has been updated.

Why? Whatever would lead you to that conclusion? People who take no great care with the first pass of a product don't suddenly wake up and become fantastic engineers.

[Kjelt]

Why? Whatever would lead you to that conclusion? People who take no great care with the first pass of a product don't suddenly wake up and become fantastic engineers.

As we have seen with the Toyota embedded software inquiry and investigation, it was a mess!
It is very hard and costly to maintain large software stacks and keep them decent over the years, from the 8 large companies I have worked for it is more an exception than a standard. It starts all nice and neat and architecture is ok and documentation ok, then 5 years later none are up to date, too many temp sw engineers have done their "job" and things start to fall apart.

[DougSpindler]

Have you read any of the accident reports where a self-driving car resulted in a fatality?  In every case so far the human would have avidoed the accident.

You seem to be under the illusion that a zero-accident rate should be attained before deploying these.

That's never happened before with any new tech, why should it happen now? SO far they seem to be doing quite well.

  The best documented accident of where the machine failed and the human driver was much better is the single car accident where the self-driving car ran into a tree killing the driver in California.  The driver had documented and reported the problem on a stretch of road where the car set-driving controlled car continually fails to stay on the road.

I'm sure the software has been updated.

There are so many simalarities between the Therac 25 and waht's going on with selfdriving cars. 
Did you not read what happened?

With the Therac when the humans were in conrtrol of the machine there were no deaths.
But with the Therac 25 the machine was automated and humans no longer had control, and deaths occured.  Society and the courts found the people the machine was supposed to be "saving their lives" wound up killing was not acceptable.


Or pehaps you want something a bit more modern and car related.  Toyota had multiple bugs in their softwsare. 


And you do know of the five times computer systems reported to the humans nuclear missles had been launced by either the US against Soviet Union/Russia?  Or that US was being attacked by the Soviet Union.

In one case it turned out to be the failure of a $0.49 chip.  In another it was the sun hitting some clouds.

And I hope you know about the bug in the Boeing 787.  Thankfully this was detected before one loaded with passengers fell out of the sky and crash landed on a city.  It was just a minor bug,,,,,  Every 248 days or 2^16 (as I recall) all of the computers would reboot.  During the boot process all displays and flioght controlls would be inopperative.  Doesn't matter if the plane was parked, in-flight, taking off or landing.

Amd yes like the Therac, defence computers, Toyota and Tesla's software, the Boeing software was in services indangering the public.

Question still remains with autodriving software.  Why is it the the programmers still mistake a tree and parked cars for the road?  This hasn't happend once, but multiple times.  Even someone dunk or high on drugs wouldn't make that mistake as often as the selfdriving car software.

I think the latest statistics say a selfdriving car would get into an acccident every 21 days if there were not a human in the car.  Imagine if all 16 miollion cars in the US were selfdriving.  Every 21 days there would be 16 million acciudents.  Car repair compnaies would love it.   




 

 














 




Did you not learn the story of the Therac 25 in school?

 

[paulca]

The "specific set of circumstances" is driving on a slippery surface. There's not many variants on that and computers already do it better than humans. The computers are so good that we've already passed laws mandating their installation to help humans drive better.

Funny you say that, but the manual of the last two cars I have owned have stated the traction control should be disabled in snow and ice as it get's confused and may prevent the vehicle from moving.

More worrying it also points out that the anti-lock brakes may cause poor braking performance on snow and ice, but they cannot be disabled.

[Fungus]

More worrying it also points out that the anti-lock brakes may cause poor braking performance on snow and ice, but they cannot be disabled.

 

Who wrote that? The Stig?

I'm sure a normal human would be worse.

[gildasd]

The "specific set of circumstances" is driving on a slippery surface. There's not many variants on that and computers already do it better than humans. The computers are so good that we've already passed laws mandating their installation to help humans drive better.

Funny you say that, but the manual of the last two cars I have owned have stated the traction control should be disabled in snow and ice as it get's confused and may prevent the vehicle from moving.

More worrying it also points out that the anti-lock brakes may cause poor braking performance on snow and ice, but they cannot be disabled.

What car are those? Audi 80’s?

The traction control and ABS on my Citroen C4 worked quite brilliantly last winter.
It does nothing at standstill (apart from locking the diff apparently) and engages more and more as the speed rises.
I tried slow speed sliding on ice in an empty car park (to learn the limits) and it did not do anything more than understeer a bit... take the safeties off, no problem, it slides.

A professional driver might beat this, but i’m not one.

[DougSpindler]

More worrying it also points out that the anti-lock brakes may cause poor braking performance on snow and ice, but they cannot be disabled.

 

Who wrote that? The Stig?

I'm sure a normal human would be worse.

Do you have any data points to support your beliefs?

We have provided many data points which have shown software is making mistakes that humans, even impaired humans would not make.

 

[Fungus]

Do you have any data points to support your beliefs?

We have provided many data points which have shown software is making mistakes that humans, even impaired humans would not make.

Yep. In your worldview the world is full of runaway cars that fail to brake and go out of control at the drop of a hat when anybody who's been outside lately knows this isn't so.

We've also shown that there's mistake so stupid that a human can't make it.

The point is this: Software con only get better as more self driving cars are put on the roads. It's like aviation, every crash leads to an investigation and usually new procedures/practices.

It's unlike aviation in that an update can be transmitted all over the world while people are sleeping.

In a few years from now cars will be far safer than humans. We just need to get over the initial hurdles.

Yes, some people will be killed by self-driving cars today but the future savings will be worth it. That's where your blind spot appears to be.

[JohnnyMalaria]

[It's unlike aviation in that an update can be transmitted all over the world while people are sleeping.

Well, we all know what forced pushed updates can do. I'm not just talking laptops and phones.

My wife was driving her 3 month old MINI on the highway at 65mph when the engine cut out. She managed to get to the shoulder. She called me to get our AAA info. I said half-jokingly that BMW must have pushed a firmware update. Well, she called the dealership while waiting for the tow truck. Yup, that's exactly what happened. They bricked her car at 65mph.

[Marco]

Modern cars are getting a little scary, EDRs, multiple always on cellular modems, a million buffer overflows which the NSA either knows about from reverse engineering or which their undercover moles put in themselves. Any advanced actor can trivially crash a modern car at will at this point with nothing more than some cellular messages, also track you and rip all the EDR data. If they don't already have a database of exploits (like the NSA) they'll just have to spend some money and time. The only certain things in life are death, taxes and exploitable buffer overflows in any non trivial C code.

I wonder if Musk rips out the antennas of cellular modems in his car, just like Zuckerberg tapes over the camera of his laptop.

[DougSpindler]

Do you have any data points to support your beliefs?

We have provided many data points which have shown software is making mistakes that humans, even impaired humans would not make.

Yep. In your worldview the world is full of runaway cars that fail to brake and go out of control at the drop of a hat when anybody who's been outside lately knows this isn't so.

We've also shown that there's mistake so stupid that a human can't make it.

The point is this: Software con only get better as more self driving cars are put on the roads. It's like aviation, every crash leads to an investigation and usually new procedures/practices.

It's unlike aviation in that an update can be transmitted all over the world while people are sleeping.

In a few years from now cars will be far safer than humans. We just need to get over the initial hurdles.

Yes, some people will be killed by self-driving cars today but the future savings will be worth it. That's where your blind spot appears to be.

Did you not read about the Therac 25 or the Dreamliner 787? 

At this statge in development of selfdriving car software can you explain why trees and parked cars are still be mistaken for roads and killing people?

[DougSpindler]

The math of why EV cars can never be mainstem.

https://youtu.be/lSb5xas_xp0

[GeorgeOfTheJungle]

LOL, for Tesla all hype is good, the fanboys might even say that this is the thing:

[Fungus]

[It's unlike aviation in that an update can be transmitted all over the world while people are sleeping.

Well, we all know what forced pushed updates can do. I'm not just talking laptops and phones.

My wife was driving her 3 month old MINI on the highway at 65mph when the engine cut out. She managed to get to the shoulder. She called me to get our AAA info. I said half-jokingly that BMW must have pushed a firmware update. Well, she called the dealership while waiting for the tow truck. Yup, that's exactly what happened. They bricked her car at 65mph.

If that's true then this will have lead to a new procedure/practice at BMW.

nb. I'm not sure that is the reason. It could just as easily be a loose wire or something, car dealers aren't famous for rigor.

[GeorgeOfTheJungle]

OTA updates means we'll all become beta testers then?

[wraper]

Did you not read about the Therac 25 or the Dreamliner 787? 

At this statge in development of selfdriving car software can you explain why trees and parked cars are still be mistaken for roads and killing people?

Can you explain why human drivers still hit trees, parked cars and kill pedestrians at this point of time?

[GeorgeOfTheJungle]

At least we always see the trailers.