First Tesla Autopilot death

[tszaboo]

Put them in the driver's seat of an 18 wheeler for 3 months and see if they are a saint.

No thank you. I also don't clean toilets after others, dont inspect tickets or teach children, fly a commercial airplane or do a brain surgery. Guess what? other jobs can be stressful. Even mine is stressful sometimes, you know, deadlines and mismanagement. On the other hand I don't start braking laws and endanger (or outright kill) people when I'm stressed out or tired.

[Someone]

What I see in this incident is that it took 130 million miles before somebody, somewhere, FINALLY found a corner case that caused the Tesla autopilot system to not react when it should have.  As a result, the entire industry has been made fully  aware that this can happen, and as a result it will likely never happen again.  A system that was already safer than manual control will become safer yet.

The rate of fatalities per 100 million km is around 0.5 globally:
https://bitre.gov.au/publications/ongoing/files/International_Comparisons_2013_Web.pdf
So on just those numbers its hard to prove that its safer than a human driver.

[System Error Message]

to those hating truckers at least watch top gear about it, they have it difficult (despite all the jokes made). There are 2 types of truckers, independent or hired (i.e. dedicated as part of a company).

It seems like no matter you go asians still drive terribly. here in the UK the brit drivers will always stop for me to cross the road but the asian drivers wont. In asia they dont even stop at the zebra crossing for you to cross not even if its a place like the hospital.

Lets hope whoever made the autopilot software wasnt asian.

[Brumby]

I want my 'Autopilot' vehicle to safely deal with someone else being a dickhead.

Unfortunately this will not happen in our lifetimes.  The only way is for the dickhead to also have Autopilot and that he cannot override it.  So no human interaction.

Now that we are back on topic, I would like to clarify this a bit more.

When I say 'safely deal with' I mean to be able to take the best option out of all the actions possible to minimise the risk to life and limb - with the objective that an accident could be avoided.  If the physics of the situation preclude that outcome, then that the injuries are as minor as possible.

I do realise that "the best option out of all the actions possible" is not a simple matter - but it should be a key objective, IMHO.

[cimmo]

cimmo, your views have been well documented in this thread, and are starting to go off topic.

Ah, another one who wants to control the discussion.
Nope, not your forum.
All I am doing is putting some perspective in to this discussion - when other posters say uninformed rubbish like "this accident wouldn't have happened if the Tesla driver was paying attention" which completely ignores the elephant in the room - the truck driver's actions (and apparent lying after the event - which most people seem to take on face value - especially if the media reports it as "fact"), then I will correct those flawed characterisations of this event.
Then I have some clowns trying desperately to find some reason why the truck driver was as innocent as the falling snow.
Why should I let those opinions go unchallenged?
Those other posters were just as "off topic" as I, so as the self declared forum cop, why did you remain silent then?
Also, when another poster declares that there is a LAW that enables a truck to do as this truck did, should I take that as granted, or am I permitted to ask for documentary evidence of this alleged "law"?

Apparently almost everyone here is quite OK with literal shitposts like "Elon Musk is a PIECE OF SHIT!"
Why did YOU let that slide?

I am not the one advocating that the rule of law is optional, I'm not the one advocating that the biggest and deadliest vehicles on the road AUTOMATICALLY have the right of way and I'm not the one saying threatening and anarchic rubbish like "yield or die" (if that poster drives a truck with an attitude like that, he needs to have his heavy vehicle licence revoked).
But apparently, you don't care when other people on this very thread want the public roads to resemble the set of a Mad Max movie.

Anyway, since there ARE some people on this thread who will need to assuage their bruised ego by having the last word, I shall leave this thread so that these other off-topic snarks, ad-homs and probable insults can happen uncontested, but I will leave with these final comments -
The fact of the matter is that this crash was NOT caused by the Tesla, it's systems or the Tesla driver (or even that piece of shit, Elon Musk). Those factors may have been contributory - maybe not, but we simply do not know how close the Tesla was to the truck when the truck commenced the turn, nor how fast the truck was going when it commenced its turn and therefore we do not know how much time was available to the Tesla's driver or its automatic systems to avoid or reduce the inevitable. I can fully visualise a nearly identical scenario involving some mundane car being driven the old fashioned way (and quite possibly at a much higher speed than the Tesla, since the Tesla is not capable of significant speeding in autopilot) resulting in an identical outcome.
Or probably a lot worse.

And for all we actually know about all those distances and timings I have just mentioned, even a perfectly operating self-driving car (without the sensor/software limitations of the current Tesla) could still have impacted this truck with enough speed to result in the death of the occupants. Perhaps slightly slower - but is severity of death a criterion worth measuring?

No, we do not need to have automatic cars be able to (magically?) avoid edge case scenarios like this - we need automatic TRUCKS to prevent scenarios like this. Prevention is always better than a cure. The sooner that obviously overworked and underpaid poor diddum truckers are put off the road and replaced by machines that are programmed to follow the rules - the better. Especially owner drivers (such as the driver of this truck) who have a documented history of bending or flat out ignoring the rules because they 'need' to. But apparently having that opinion makes me a 'trucker hater'.

If truckers (such as Mr Baressi) insist on driving like entitled arseholes, then they deserve the hate - and need to be replaced with tech that knows the road rules and is programmed to adhere to them - even if this means the robo-truck has to wait all day until the road is clear of traffic that has the right of way.

[Brumby]

cimmo, your views have been well documented in this thread, and are starting to go off topic.

Ah, another one who wants to control the discussion.

[GEuser]

The FUture .

All truckies all taxi's fully automated , now off to find them another job .

[iampoor]

cimmo, your views have been well documented in this thread, and are starting to go off topic.

Ah, another one who wants to control the discussion.
Nope, not your forum.

Never made either of those claims. I posted this thread, and was hoping to gain a better understanding of the possible failings of the technology and potential legal implications of this accident. Many posts were very helpful and educational. 

Your posts on the other hand, are arguing something that the majority of us agree with (Yes, we know the truck driver most likely should have not turned). It doesnt matter. He did. This is a practical issue that must be solved with these systems. Your point makes sense, but it doesnt have anything to do with what already happened.

Coming into a thread with a poor attitude, shoveling shit on a group of people you apparently dont like is a great way to get a thread locked. If you want to rant about your issues with truck drivers, make a new thread.

 

[suicidaleggroll]

What I see in this incident is that it took 130 million miles before somebody, somewhere, FINALLY found a corner case that caused the Tesla autopilot system to not react when it should have.  As a result, the entire industry has been made fully  aware that this can happen, and as a result it will likely never happen again.  A system that was already safer than manual control will become safer yet.

The rate of fatalities per 100 million km is around 0.5 globally:
https://bitre.gov.au/publications/ongoing/files/International_Comparisons_2013_Web.pdf
So on just those numbers its hard to prove that its safer than a human driver.

Those results are misleading, because their "average" of various countries was not weighted by distance traveled.  They weighted iceland's rate of .54 with only 3185 km traveled as equal to the US's rate of .68 with nearly 5 million km traveled.  When you total the deaths and the distance traveled to get the actual average, you get a rate of .65 deaths per 100M km, which comes out to 1 death per 95M miles, which is comparable to other results I've found.

[zapta]

What I see in this incident is that it took 130 million miles before somebody, somewhere, FINALLY found a corner case that caused the Tesla autopilot system to not react when it should have.

This is an impressive stats and is better than human drivers' stats but it may also be biased.  The Autopilot (and I presume similar technologies from other companies) is used on 'easy' miles with good weather conditions, non too dense traffic, highways stretches, etc, and with well maintained expensive new cars.

To do a proper comparison with human drivers, the other variables must be controlled.

[suicidaleggroll]

What I see in this incident is that it took 130 million miles before somebody, somewhere, FINALLY found a corner case that caused the Tesla autopilot system to not react when it should have.

This is an impressive stats and is better than human drivers' stats but it may also be biased.  The Autopilot (and I presume similar technologies from other companies) is used on 'easy' miles with good weather conditions, non too dense traffic, highways stretches, etc, and with well maintained expensive new cars.

To do a proper comparison with human drivers, the other variables must be controlled.

That is true, it's hard to make an accurate comparison when the autopilot is only used in certain conditions and only with certain vehicles.  I do challenge your argument of them being "easy" miles though.  While yes it should be safer since it's in good clear weather, my gut tells me that's actually when more fatal accidents happen.  After all, when the weather is poor, visibility is bad, or traffic is dense, the average vehicle speed is significantly slower.  So while the risk of an accident might be higher, I'd argue that the risk of a fatal accident would potentially be lower than when the conditions are clear.  After all, every fatal accident I've seen or heard of in my local area (as far as I can recall) has always been when the sun is shining and the birds are chirping, and somebody does something stupid.  There are plenty of fender-benders when the snow comes out, but nobody is driving fast enough for serious injury.

What's actually more impressive to me though is not the fatal accident rate, but the overall at-fault accident rate.  In the US, as far as I can tell the average accident rate is about one per 200-500k miles (some sources say 160, some say 250, some say 500, let's go with 300k).  Assuming the average number of vehicles involved in each crash is 3 (it's probably closer to 2, but this will make the numbers more conservative), one of which is at-fault, that means on average there is one at-fault accident per million miles or so.  The Tesla autopilot has racked up 130M miles, and as far as I can tell from the research I've done, still has zero at-fault accidents.  It's only on the highway, sure, but still impressive.

[SkyMaster]

...  The Tesla autopilot has racked up 130M miles, and as far as I can tell from the research I've done, still has zero at-fault accidents.  It's only on the highway, sure, but still impressive.

This one came out yesterday:

Could this be the first?

Second Tesla autopilot crash under federal scrutiny
http://money.cnn.com/2016/07/06/autos/tesla-autopilot-accident/


The same article is publishing the following:

Federal data shows that human error is responsible for 94% of the 35,200 traffic fatalities last year. That works out to 1.12 deaths for every 100 million miles driven.

Tesla says its owners have driven more than 130 million miles using autopilot. The one fatality works out to 0.78 fatalities per 100 million miles.

[suicidaleggroll]

...  The Tesla autopilot has racked up 130M miles, and as far as I can tell from the research I've done, still has zero at-fault accidents.  It's only on the highway, sure, but still impressive.

This one came out yesterday:

Could this be the first?

Second Tesla autopilot crash under federal scrutiny
http://money.cnn.com/2016/07/06/autos/tesla-autopilot-accident/

I saw that too.  Another article I was reading on it said the guy claimed he was using autopilot, but the detailed logs from the car weren't transmitted to Tesla after the accident, and they've tried to call him three times with no answer or reply, so nobody really knows yet.

[SkyMaster]

  The Tesla autopilot has racked up 130M miles, and as far as I can tell from the research I've done, still has zero at-fault accidents.  It's only on the highway, sure, but still impressive.

The fatal crash from May 7th, there was an obstruction on the road (at this point it does not matter if it is a truck or a dead cow).

Did the Tesla "autopilot" applied the brake?

Did the Tesla "autopilot" tried to stop or veer off course?


If the Tesla "autopilot" kept the car on course and did not try to stop; I would presume that the Tesla "autopilot" is at-fault in this particular case.

[Brumby]

  The Tesla autopilot has racked up 130M miles, and as far as I can tell from the research I've done, still has zero at-fault accidents.  It's only on the highway, sure, but still impressive.

The fatal crash from May 7th, there was an obstruction on the road (at this point it does not matter if it is a truck or a dead cow).

Did the Tesla "autopilot" applied the brake?

Did the Tesla "autopilot" tried to stop or veer off course?


If the Tesla "autopilot" kept the car on course and did not try to stop; I would presume that the Tesla "autopilot" is at-fault in this particular case.

These are the type of questions that need to be asked.... and answered.

[fubar.gr]

So rather that willingly break the law - to fail to yield to traffic that has the right of way, perhaps the truck driver in that scenario should continue along the road until he finds a controlled intersection - a set of traffic lights - or perhaps a roundabout - which will enable him to turn left LEGALLY and SAFELY?

What I described in my previous post is neither necessarily illegal or unsafe and doesn't count as failure to yield, as long as the other car is far away enough to slow down safely.

In that case, the law says that you have to yield for vehicles already in the intersection, even if you have the right-of-way.

You seem to operate under the assumption that as long you have the right of way, no one else should even dare to inconvenience you by getting in your way, even if this is done in a safe manner, and all you have to do is slow down.

It doesn't work that way. Roads are a shared commodity and everyone has a right to use them. You must always drive in a manner that allows slower vehicles, cyclists, pedestrians, etc to share the road with you.



So the real question is this: Did the truck cut the Tesla off? Or was the truck for long enough in the intersection for the Tesla driver to see it and slow down?

This is where the point of impact matters. If the Tesla had hit the front of the truck, then one could argue that the truck cut the Tesla off and there was not enough time to avoid the accident.

But the Tesla hit the truck at the gap between the trailer wheels.



That's right in the middle of the whole truck length. Trucks aren't particularly fast, so that means it was for quite some time in the intersection. In that case the Tesla driver had to yield.

[nctnico]

...  The Tesla autopilot has racked up 130M miles, and as far as I can tell from the research I've done, still has zero at-fault accidents.  It's only on the highway, sure, but still impressive.

This one came out yesterday:
Second Tesla autopilot crash under federal scrutiny
http://money.cnn.com/2016/07/06/autos/tesla-autopilot-accident/

Federal data shows that human error is responsible for 94% of the 35,200 traffic fatalities last year. That works out to 1.12 deaths for every 100 million miles driven.

Tesla says its owners have driven more than 130 million miles using autopilot. The one fatality works out to 0.78 fatalities per 100 million miles.

This is a fine example of statistics turned into lies. First of all you need to break down on what kind of roads (inner city, rural, highway, etc) the autopilot has been used and compare statistics for each type of road to human drivers versus autopilot. In general the fatalities per distance on a highway are 20 times LESS than inside a city. So if the autopilot is mostly used on highways this will bias the statistics heavily in favor of the autopilot.

[Brumby]

First of all you need to break down on what kind of roads (inner city, rural, highway, etc) the autopilot has been used and compare statistics for each type of road to human drivers versus autopilot. In general the fatalities per distance on a highway are 20 times LESS than inside a city. So if the autopilot is mostly used on highways this will bias the statistics heavily in favor of the autopilot.

Perhaps drive time might be a more appropriate measurement.

After all, as you say, a mile in city traffic represents a much greater risk than a mile on a clear highway - but an hour on each would, I suggest, provide more comparable risks.  Speed then becomes a simple parameter, rather than a fundamental component in the definition of the frame of reference.

In Australia, we have a "Stop. Revive. Survive" campaign that gets rolled out, especially during holiday periods.  One of the basic recommendations is to take a break every two hours.  Although implied, they don't make any specific distinction whether it's city traffic or open freeway.

[nctnico]

The official statistics work per distance so that seems to be the agreed upon measure.

[Brumby]

Doesn't make it the best measure.


It is, however, the easiest ... which provides a convenient and consistent measure - which is important in itself.

[dannyf]

Federal data shows that human error is responsible for 94% of the 35,200 traffic fatalities last year. That works out to 1.12 deaths for every 100 million miles driven.

Tesla says its owners have driven more than 130 million miles using autopilot. The one fatality works out to 0.78 fatalities per 100 million miles.

Whether such a comparison is valid depends on what you are trying to compare. Two factors should be considered:

1) any fatality is a combination of vehicle safety + driver behaviors. Take Volvo cars for example. Because of its safety reputation, safety conscious drivers tend to buy them. Those drivers are less aggressive and are less likely to get involved in accidents -> lower fatality. When you condition the fatality rate on the number of accidents those vehicles are involved in, Volvo actually has one of the highest fatality rates -> it doesn't do a good job in protecting its occupants -> true measure of safety.

In this case, comparing fatality rates without controlling other factors will assume that the drivers are the same (I would argue that tesla drivers tend to be less aggressive than the average driver on the road), driving conditions are the same (I would argue that the auto pilot is engaged more likely in open high way driving -> safer than in-town driving), vehicles are comparably maintained (I would argue that given their newer age and more affluent ownership, a tesla is better maintained than average cars on the road), ...

2) confidence level / sample size / unobservable variables: it is hard to comparable two statistics from two vastly different sample size: 1 tesla vs. hundreds of thousands of other vehicles over decades of operations. The confidence range for the tesla fatality rate must be pretty wide, rendering the point estimate unreliable.

[fubar.gr]

This one came out yesterday:

Could this be the first?

Second Tesla autopilot crash under federal scrutiny
http://money.cnn.com/2016/07/06/autos/tesla-autopilot-accident/


The same article is publishing the following:

Federal data shows that human error is responsible for 94% of the 35,200 traffic fatalities last year. That works out to 1.12 deaths for every 100 million miles driven.

Tesla says its owners have driven more than 130 million miles using autopilot. The one fatality works out to 0.78 fatalities per 100 million miles.

This doesn't instill much confidence. According to these numbers, the autopilot is marginally better than a human driver.

Also, as nctnico points out, if those miles are mostly done on highways, where accedents are an order of magnitude fewer, this makes the Autopilot actually perform worse than a human driver.

On the other hand, since there are too few Teslas on the streets, statistically speaking these numbers are meaningless. The Autopilot could be much much safer in reality than a typical human driver. Only time will tell.

My only fear is that regulators might have a knee-jerk reaction and axe this system completely. This would be a pity because it looks really promising.

[nctnico]

I'm very sure auto pilot systems will not be axed. Even before Google even thought about making a self driving car many research institutes have been experimenting with this technology for decades. Over a decade ago I worked for such an institute and experienced a demo of a car in which you only had to stear. It did throttle and brakes itself based on data coming from the cars in front.

The biggest advantage is that self driving cars can talk to eachother and get data from sensors in the road. The way we humans drive is insanely inefficient and bad. If you sit next to a highway and watch the traffic you'll notice traffic flow isn't continuous but cars come in (accident prone) packs. Just look at how some people like to drive behind another car so they have a tail to follow (just like elephants do). Also one slower driver is enough to cause a major traffic jam during rush hour. Not to mention the utter chaos caused by an accident. All in all autonomous cars are very necessary to increase the throughput of the infrastructure.

Still you can argue whether Tesla introduced their system too early or has been too confident people wouldn't abuse it.

[bitslice]

When you condition the fatality rate on the number of accidents those vehicles are involved in, Volvo actually has one of the highest fatality rates

I find that hard to believe (source?)
The Volvo XC90 has a zero death rate, and both Volvo and Saab are really heavily built cars.

The consistently get top results in crash worthiness, so to suggest that such a metric is misleading is odd.

[nctnico]

Volvo's have not been safer or better than any other car for a long time. Volvo has been owned by Ford so you are basically buying a Ford with a Volvo badge.