"The driver had received several visual and one audible hands-on warning earlier in the drive and the driver’s hands were not detected on the wheel for six seconds prior to the collision."
Shouldn't you have more audible warnings than visual if you are aware that not complying will lead to an accident?
This feels particularly evasive. If you want to claim that your technology is 10x safer, you should be able to back that up with self-insurance, and not try to put the blame on the driver.
If the autopilot does not think it is capable of driving safely, it should pull over to the shoulder of the road, not beep and flash.
It doesn't seem like pulling to the right would be safe in this situation. I think that automatically slowing down would be, though. Even slowing at a rate similar to a human letting their foot off the gas (to avoid unexpected behavior) would reduce the chance of fatality in a crash.
People do not pull over to the shoulder when they are between lanes in (presumably) busy traffic, which is where you have to be to hit a median.
I am not sure if Tesla's current system, which is still just cruise control with steering, should ever be changing lanes unprompted. I don't have a Tesla but I would not want my car's cruise control to do that right now. I do think that pulling over in safe conditions is something that a self-driving system needs to be able to do at some point on the path to full autonomy.
In this situation I think the lane to the right is not the shoulder. So in a situation where there are only seconds to respond to an obstacle, there might be cars blocking the Tesla on the right.
That's very troubling. If you saw the Uber crash cam on the driver it's obvious that the "responsible drivers" are completely non-responsive, on their phones, half asleep. A system that lulls the driver into a false sense of security does NOT pass responsibility to said driver.
I'm guessing that if I buy a Tesla with autopilot, I'm not going to find any insurers who will charge me discounted rates that reflect autopilot being 10x safer.
This is a blatant excuse - you can gently hold the wheels ready to take over without applying pressure and their system will not detect it you don't firmly hold the wheels, which defeats the comfort of having bought "Autopilot" in the first place.
Actually the warning had nothing to do with the crash. if i drove on AP for 15 minutes and i was alerted to put my hands on the wheel 2 minutes in. It would fit tesla's specially crafted statement.
making it look like user was warned and didn't respond by putting their hands on the wheel but like i previously said, its completely false.
> Tesla Autopilot does not prevent all accidents – such a standard would be impossible – but it makes them much less likely to occur. It unequivocally makes the world safer for the vehicle occupants, pedestrians and cyclists.
These kinds of unequivocal statements are just begging for a lawsuit, imo. They simply don't know that this is true.
Exactly. They are taking both sides of the an argument to get credit but not take blame. They want credit for allegedly reducing crashes, but they don't want to be held responsible for driving into a wall.
> In the US, there is one automotive fatality every 86 million miles across all vehicles from all manufacturers. For Tesla, there is one fatality, including known pedestrian fatalities, every 320 million miles in vehicles equipped with Autopilot hardware. If you are driving a Tesla equipped with Autopilot hardware, you are 3.7 times less likely to be involved in a fatal accident.
Let's put aside the consideration of how the average Tesla driver differs compared to the average American driver, or how the average mile driven by Autopilot differs compared to the average mile driven by all human drivers; I'm a little bit bothered by the language here:
> If you are driving a Tesla equipped with Autopilot hardware...
> every 320 million miles in vehicles equipped with Autopilot hardware...
Why is the language "equipped with Autopilot hardware" used? Tesla isn't able to give an estimation of when Autopilot was equipped and engaged?
That's an interesting question. Similarly, one could ask why the focus on fatalities rather than crashes as Tesla's superior traditional safety rating means fewer crashes should be fatal, all else equal.
However, these questions don't lead anywhere interesting. Model S and X cars were driven before and after Autopilot was activated (no hardware change), and the crash rate was reduced after activation. This is the "40% reduction" mentioned by Tesla in the article you pulled the quotes from. You can read a summary here: https://www.theverge.com/2017/1/19/14326258/teslas-crash-rat...
This 40% has nothing to do with Autopilot and applies to any cars with auto braking.
The Insurance Institute for Highway Safety says automatic braking — which can bring the car to a stop if it senses an imminent collision — can reduce rear-end crashes by 40 percent.
My bad -- I thought that was one of the features of Autopilot. It still looks like the safety improved after the Autopilot rollout, but hopefully some public numbers with more granularity in regards to these software features can be released.
Tesla likes to market their version of automatic braking as part of Autopilot because it makes the safety of Autopilot look better, but it's a fairly common feature that is available on many models of cars, that the entire industry plans to make standard on all new cars in a few years, and that operates even when Autopilot mode - the controversial feature that drives Teslas into concrete barriers and trucks - is not engaged.
Also, they've had 3 fatalities so far. If one fully occupied model x (5 people) crashes and kills everyone... it's back to the average. If the people in the other car die too, it's less than average.
That safety advantage can be wiped out in a single accident. Tesla has been lucky that all 3 have been occupied by a single person.
In the moments before the collision, which occurred at 9:27 a.m. on Friday, March 23rd, Autopilot was engaged...
The rest may be true, may be germane, or may not be. The only part that matters to me until the results of the NTSB investigation is made public, is that the autopilot did this.
It's more about what the autopilot didn't do than what it did do. It did drive straight ahead into a concrete barrier. It didn't see the barrier. If it had seen the barrier (as is reasonable to expect), it wouldn't have driven into it. Thus what it didn't do caused what it did do.
The fact that Tesla even made a press release of this nature disturbs & annoys me.
When the NTSB is investigating an aircraft or train accident we don't see the vehicle manufacturer rushing out a "well we told him to take control" statement and a dump of telemetry. At most they state "we are of course working with the authorities to establish the cause and any corrective actions" until the investigation is completed.
Tesla needs to mature from a blame-the-user software-industry mentality to something more befitting their responsibilities.
Really not much additional info here aside from the number of warnings the driver had. Also, the standard "Tesla is 10x safer" metrics that get pulled out each time a crash gets sensationalized.
What I think they fail to address, especially in this case, is that the autopilot did something a human driver who was paying attention would never do. Autopilot does a great job of saving people from things even a wary driver would miss, much less a negligent one, but the fatal accidents that occur in statistics are not from people who are fully watching the road missing the fact that there's a concrete barrier with yellow safety markings directly in their path, and hitting it head on for no good reason (e.g. evasive action because of another driver, etc, oh a stupid last-minute "oh crap that's my exit").
I want autopilot to succeed, and I want Tesla (and Musk) to succeed, and for the sake of their public image they have to realize that this isn't an average accident statistic, a lapse in attention or evasive maneuvering. It's a car that seemingly plowed right into a concrete barrier while still under complete control. That's not a mistake a healthy human will make.
> the autopilot did something a human driver who was paying attention would never do.
Then how did the crushed barrier get crushed before the Tesla hit it? Clearly, the stretch of road is unsafe enough to trick humans drivers (and, clearly, Tesla should improve).
I'm obviously not privy to the details of the prior crash, but I'm pretty sure any healthy human would not deliberately drive into the barrier.
The most likely scenarios I can think of would be not paying attention and drifting into the barrier, attempting to avoid a car merging into the lane (and not paying enough attention), or being struck by another car and being forced into the barrier.
At that particular point, what I see the most is people deciding that they either need to exit or not and abruptly crossing from one HOV lane to the other. Lots of crazy shit. So I'm not sure your scenarios have much relevance -- how many times have you driven through this intersection? I've been through it maybe a thousand times since it was built.
I’ve seen people back up in the breakdown lane on the expressway because they missed their exit. But that was Beijing, and all drivers are hyper sensitive on the lookout for that kind of stuff.
On my first drive into Bucharest a guy drove on the wrong side of the divided highway to take a left into the lot of a building supplies company right in front of my car. Scared the crap out of me.
Can confirm, have just driven for 2+ hours on the streets of Bucharest and there’s no way for an AI-like car to drive in here for more than 2 blocks without getting involved into an accident. Unless said car becomes sentient, but even in that case it would still have to “guess” some of the crazy stuff the other drivers on the road may be up to or to expect that a barely marked road-construction thingie will suddenly show up just in front of the vehicle while driving at 60 kph. And Bucharest has got nothing on a city like Istanbul.
It's the only place where I got honked at by a car on the sidewalk while walking on the sidewalk, and when I didn't get out of the way fast enough he proceeded to attempt to nudge me with his car.
I think the answer is that the path into the barrier looks just like an actual lane, and that this was enough confusion that a human driver apparently had made the same exact mistake a week earlier.
https://imgur.com/a/iMY1x
The lanes aren't very well indicated, but given how visible the divider is, I don't think either an autopilot or a driver should drive off.
Reminds me of a scene in "La grande vadrouille" where a motorcyclist is killed on a mountain road because he was following the dotted line and the painter had gone to the side of the road to take a break.
This divider looks rather unforgiving. Is it common for dividers in US to be just plain concrete blocks without anything to prevent rapid deceleration?
that image also makes me wonder if tesla's system is designed to identify knocked-over cones. It's one of those things that would be easy to code for, but also easy to overlook.
It's quite hard if you take into account that machine learning needs 10000x the training data that a human needs, and Tesla is sold in 30+ countries (and used in much more).
I think 3d mapping the surrounding world correctly using multiple cameras (what humans are doing) is more generalizable.
That still looks pretty risky to me - I compared a similar junction here in the UK near to where I live (on the M90 in Fife) and it has about 40 impact attenuation barrels in a triangle.
Those are also present in many locations in the US, although they seem to be phasing them out so I think they are considered old tech. I don't know the performance specs, but new cars are a lot smaller and safer than old cars - the same is probably true here too.
Except those solid white lines delimit an uncrossable barrier delimiting the HOV-specific (during commute hours) flyover from the 101 lane, which is well documented for nearly a mile preceding that point.
Reasonable, lawful, absolutely. People break the law, people get drunk, people get sleepy.
There's a spot just like this in Houston on 610 East. HUUUUUUGE flyover as the HOV lane spends about a half mile merging into the left lane of 610 (nice fat fast freeway, they need the runway). A guy had thought the flyway would be a safe place to park his car and wait for a tow. Safe enough that he was sitting on the hood of his car. Sure as shit the Challenger in front of me got confused, cut into the flyaway thinking it was a lane, hit the disabled car, and sent the guy flipping up into the air a good 10, 15 fit. Fucked up accident.
Anyway point is after the accident I was talking to the driver of the Challenger, trying to calm him down (kept saying "holy shit I fucking killed that guy!") and other that shock he was sober as a duck. He got bloodtested and everything, clean. In court he said he just got confused and thought it was a lane.
LONG STORY SHORT human error mang, human error. Still not sure why a car was able to do it.
They have those same white lines leading into intersections and people cross that shit all the time. People speed.
If you are building an autopilot system, I’d expect 1) identify black/yellow hazard signs and 2) don’t hit them to be basic features by version 0.5. Seems like a big miss to run straight into that.
Maybe SDC already do this, but it seems like the Agent should keep track of things that look like potential lanes and then if the potential lanes turn out to be not-lanes (have barrier in them) the Agent should flag the area that looked like the start of a lane so future SDC know to avoid it.
The entire thing is two solid white lines on either side, which are pretty strong indicators that you shouldn't pass into that area in the first place, diagonal stripes or not.
It also indicates that you shouldn't pass out of that area.
The vehicle should have at the very least stopped short of the road obstruction. But we don't yet know what led up to being in that lane, I was hoping the article in the OP would have taken us through it. Instead I'm getting the feeling this was a catastrophic software failure. Without more transparency on the issue, despite what Tesla may prefer I don't think I have any choice but to feel uncomfortable.
I wonder if the problem is the uneven pavement leading up to it. The white solid line starts close to it, then drifts left. In certain lighting conditions, the division in the pavement might seem more prominent than the white line, which I could imagine tricking a human or computer to follow it instead. That wouldn't cause a head-on crash itself, but would cause the car to get very close. I could then imagine a computer trying to correct its mistake by moving RIGHT (because that area looks like a lane) instead of left and going head-on into it.
Though of course that doesn't explain why it didn't recognize the barrier as something it should avoid at all costs. Unless perhaps something to the left confused it and made it think there was an obstacle there, too, thus causing it to think it was going to crash regardless. If so, maybe it did try to brake at the last second?
It's not that straightforward. It's two adjacent lanes, which gets separated. Left one becomes an exit, the lane next to that one continues. The space between them keeps growing. The two lanes in question have a solid line between them before they separate to indicate you can't cross between them.
any healthy human would not deliberately drive into the barrier
But plenty have driven negligently into barriers.
The worst such crash I've ever seen was barely 2 miles from there, on northbound 85 near Fremont Ave. There is a soundwall that comes to a connection point where a wall segment is directly perpendicular to the freeway. For some reason, the guardrail had a gap there almost exactly the width of a vehicle.
A few years ago, a vehicle veered off the right shoulder and perfectly threaded the gap, into the wall at full speed. It was compressed to maybe 5 feet long.
It's a somewhat easy accident to have in heavy traffic if the lanes aren't marked clearly and you don't see the barrier because of cars ahead of you. Not a factor in the Tesla crash (they had a clear line of sight) or in the original crash (which was a DUI).
it's a mistake that healthy humans make all the time, which is why those crash barriers exist in the first place, and why that particular crash barrier was damaged by a previous crash.
You’re missing the point. The crucial fact according to OP is that the car did something that a fully aware driver would not ever do. It’s at least worth acknowledging that.
"Fully aware" drivers (true Scotsmen) as proposed here statistically do not exist, or at least do not compose the vast majority of drivers. So it is kind of a meaningless comparison to draw.
It's not a True Scotsman in this context. The point is that humans tend to crash into things they're not fully aware of, short of self-harm.
Either:
A) The autopilot could not see a concrete barrier in its path.
B) More likely, and as the story reports, it WAS aware of the danger but didn't do anything.
In which bucket would you put the possibility that the computer observed evidence of the barrier, but the evidence that there was no barrier -- the prior that barriers don't exist in lanes of traffic -- was too strong?
"Fully aware drivers" are all the many drivers who manage to navigate that area without driving into the barrier, minus the ones who didn't mean to get off there or meant to get off and didn't. The latter groups are aware enough to drive safely through the area but still spacing out a little.
It sounds like peering into the black box of autopilot and anthropomorphising some parts of it but not others. Why not say "autopilot sometimes drives into barriers and also humans sometimes drive into barriers."? In both cases the driver/autopilot failed for whatever reason.
A swerve is not the only possibility, as the area between lanes looks like a lane. My father has done the same thing while driving at night. He insisted he was in a lane while I yelled repeatedly that he wasn't. Luckily he changed lanes / entered a real lane just before we crashed.
Yes, bad visibility could explain it, but Tesla themselves say it was 9.30am and the autopilot had "five seconds and 150 meters of unobstructed view of the concrete divider".
Nobody is trying to absolve Tesla. They're just refuting a dumb and easily rebuked claim that no human driver would make the same mistake. There is evidence that crashes happened there before, and it would be easy to imagine someone crashing there because they were focusing on trying to find a spot to merge while driving in what appeared to be a standard lane.
Again, not absolving Tesla, just being realistic about the capabilities of human drivers.
It's only anthropomorphizing in the sense that we're discussing two systems capable of propelling automobiles around. Presumably, they're both in the business of avoiding concrete barriers.
If one system is steering the vehicle into things that the other system would, in most conditions, reliably avoid, it bears some discussion.
I've been a fully aware driver unsure of what to do when seeing a surprise in my lane. I only hit the break because a passenger started screaming. I was so confused to see a pedestrian on the freeway.
I thought the auto pilot warned the driver to take control of the vehicle or is that incorrect? There is probably a lot more we don't know and it would be premature to place the blame on the auto pilot and Tesla. This isn't a lvl 5 system. It still requires the drivers attention and that is something we need to consider ( he obviously didn't have his hands on the wheel).
Technically tesla's statement says he was warned at some point to take control during his drive. That could have been 5 seconds into his commute, or it could have been 5 seconds before he hit the barrier. He may have taken control at that time, and then reengaged it later. He may or may not have been warned to take control before hitting the barrier.
The statement mentions "hands on warning" which just means you need to keep your hands on the wheel. That's very different than giving control back to the driver.
but the average person in the US anyway is not a good driver. i very often see people get aurprised at the lane ending at that exact spot. they should put a rumble strip leading up to it.
So, my question is, is it fair to compare AP to a human driver paying attention or is it more fair to compare it to an average driver.
I mean that just for the sake of argument. In this specific case, NTSB or someone should ban AP. It is so obvious what’s about to happen there, an AP should do what a driver should do which is take the ramp whether it’s a wrong turn or not. So many a-holes try to squeeze in last second (oops) when they should just take the damn exit or miss the damn exit as the case may be. What AP did here is what a poor and panicked driver would do and that’s just not acceptable.
But this wasn't a self driving vehicle. It isn't capable of dealing with all situations, and it warning the driver for 6 seconds to take control, which didn't happen.
Humans do exceedingly stupid things all the time because they stop paying attention, even momentarily (or subconsciously).
We put big lights on the back of cars that light up when they brake. And yet, despite a driver looking direct at a huge object with two lights, that rapidly grows larger right in front of them, does not always prevent a collision. Or even a chain of collisions. I don't get on the road much, and yet even I've seen a ton of accidents that are baffling and can only have been the result of a driver not paying attention for a bit.
I'm reminded of how in quite a few places removing signs and lights actually improved safety because it forced drivers to stay aware instead of 'driving on autopilot', so to speak.
I think that's the real issue here. The more we outsource our attention to a machine, the more important it is that said machine does MUCH better than we humans do. Especially if a mistake can be deadly.
But I wouldn't be surprised if, indeed, technically this accident could've happened just as easily by a non-autopilot car where the driver had a little 'micro-sleep', got distracted by something in his field of view, mistakenly thought he was on a lane and didn't notice the (let's be fair) ridiculously bad markers that were the only way to tell that part of the 'gray' stuff in front of him was in fact a wall of concrete.
I mean, just look at the image: https://imgur.com/a/iMY1x . Half of what makes the barrier stand out is the shadow!
All that said I might sound more argumentative than I am. I do agree with most of your comment.
«the autopilot did something a human driver who was paying attention would never do»
I'm sure that out of the 1.25 million annual automobile-related deaths, plenty of drivers were paying attention and still did stupid things similar to this accident.
Using global statistics greatly skews your argument. While all these deaths are tragic, many of them could be prevented through law and regulation of human drivers.
This statement is meaningless circular logic. You can handwave away any incident with a human driver by saying he wasn't "healthy".
A human pilot has intentionally driven an airplane full of passengers into the ground with full control because he wanted to kill himself. The airline believed he was "healthy".
I want electric cars to succeed as I am am long bored of breathing in vehicle fumes. Automated driving is cool sci-fi tech, but holds nowhere near the same sense of necessity, as far as I am concerned.
So, I thought it wasn't all that clever in the first place to try and marry the risks of getting electric cars into the market with the risks of telling the extremely wealthy that they didn't need to hold the damn steering wheel.
> In the moments before the collision, which occurred at 9:27 a.m. on Friday, March 23rd, Autopilot was engaged with the adaptive cruise control follow-distance set to minimum.
Would've loved to know the exact moments we are talking about here. Is it 5 seconds or 5 minutes?
> The driver had received several visual and one audible hands-on warning earlier in the drive and the driver’s hands were not detected on the wheel for six seconds prior to the collision.
What does this mean? Did Tesla want to give over control to driver? Or just normal no hands?
> The driver had about five seconds and 150 meters of unobstructed view of the concrete divider with the crushed crash attenuator, but the vehicle logs show that no action was taken.
Again its hard to know if Tesla wanted to give up control or not. I hope they aren't just saying that because the driver's hands were off the steering wheel, the crash occurred and is his fault for not obeying the agreement.
> I hope they aren't just saying that because the driver's hands were off the steering wheel, the crash occurred and is his fault for not obeying the agreement.
That's exactly Tesla's official stance - if you crash you're SOL and at fault, Autopilot or Auto Emergency Braking be damned.
https://www.cnbc.com/2018/01/31/apples-steve-wozniak-doesnt-...
"Man you have got to be ready — it makes mistakes, it loses track of the lane lines. You have to be on your toes all the time," says Wozniak. "All Tesla did is say, 'It is beta so we are not responsible. It doesn't necessarily work, so you have to be in control.'
As a Tesla owner, this is not accurate wording. The car will very quickly decelerate until about 5 mph if you have the crawl feature on.
I need to make a post, as a long time Tesla owner/very frequent autopilot user, because I think I know exactly what happened. Unfortunately I need to go to bed, already up later that I should be for tomorrow. A lot of information and comments I’ve seen about how it works and how autopilot is dangerous etc, are quite wrong. I think it’d be important for people to understand how it works from people who use it and get their opinions first before making statements about AP in general.
For the record, I love using AP and will continue to do so. But it is no silver bullet, and 99.5% of Tesla owners know this and how to correctly use it.
Personally I would only use a self drive feature on the middle lanes of a highway if I was going to be distracted with something else, there’s too much bullshit going on at the edges for it to be safe.
As someone who drives this exact stretch of road every day, in a Tesla, I agree. This lane is the end of a long stretch of road with either zero or minimal shoulder, where even the slightest glitch could cause a scrape on the side of your car. I'm always paranoid driving this stretch.
This concrete barrier is also quite a distance past where this left exit lane starts to separate from the rest of US-101, so there are really only two ways you'd risk hitting it. First, and probably most common, is people who change their minds last-second about taking that exit. Second, would be a serious AP screw-up (like what might have happened here).
FWIW, I've never had AP engaged when driving through this spot, and almost instinctively disable it when taking a highway off-ramp/change-over.
How much can we trust this data, provided by Tesla? Maybe the sensors that tell if the hands were on the wheel malfunctioned? Maybe he tried to restle the car into control? Would be helpful to have a video recording of the driver instead.
I would trust the data, because lying at this point could be criminal, or civilly significant. I wouldn’t trust all of the fluff and interpretation and hand-waving around the data though.
A level of skeptism is always healthy. Also recognise there are a endless 'maybes'. Anyone that works with data knows there has to be a level of trust and common sense otherwise you hit analysis paralysis.
And on 'trust' Tesla have a history of being open and upfront. Organisations tend to repeat a culture. So for Uber I would closely question their statements given an ongoing history of unethical behaviour across many levels. It seems likely they will lie or show limited truth to reach an agenda. Tesla's history seems open and honest.
Yeah, they've been open and upfront about things like the fact you can't maintain them yourself (they refuse to provide service manuals except where required by law in MA, where even then you pay a three digit fee to -view-, not get, a copy, and have to make an appointment).
Or the fact that they will remotely downgrade your car (to an older software version) and remove features of it (disable ethernet ports) if they find you "snooping" and discover references to new models.
Thats the kind of openness and honesty you laud them for?
This is 100% anecdote: A friend of mine test drove a Tesla, and said that when he was in the far right lane on a highway, the autopilot would get a little sketchy at off ramps. It didn’t ever off or anything, he described it as “jiggling” like it wasn’t sure what do when the right lane marker disappeared. He found that less than inspiring, although he did find the acceleration intoxicating.
very deceptively crafted sentence. if i drove on AP for 15 minutes and i was alerted to put my hands on the wheel 2 minutes in.
It would fit tesla's specially crafted statement.
It makes it look like user was warned and didn't respond by putting their hands on the wheel but its completely false. The AP drove him straight into the barrier with no warning.
It seems gross that they spend so much of this saying "yeah he died and it may be our fault, but statistically you're still better off as our customer."
And that's assuming they can identify that autopilot is statistically safer. If autopilot isn't used in the same conditions as all human driving, the rates aren't apples to apples comparisons, and the "you are 3.7 times less likely to be involved in a fatal accident" figure they imply is misleading.
If that's literally the best case scenario, then that's pretty damn pitiful and a gross misleading of facts using statistics. It's a shame. Let's use "AP disengage" as the equating metric to human-caused crashes and see how it stacks up.
They didn't even control for wealth. People that can afford a Tesla and the auto pilot add on are much less likely to be driving on drugs, alcohol, etc. It doesn't say anything about the statistics when auto-pilot is engaged, it just says cars with auto pilot. And it doesn't give the statistics for Teslas without auto-pilot (Teslas have other safety factors that might mean auto-pilot isn't responsible for the mortality reduction, etc.)
Also look at occupation. Tesla owners = disproportionately software engineers. Software engineers are one of the lowest car insurance risk pools (and we're even before any of them started using auto-pilot).
"The driver had received several visual and one audible hands-on warning earlier in the drive ... and the driver’s hands were not detected on the wheel for six seconds prior to the collision."
Seems like an oxymoron that's getting drivers into danger.
>Tesla said its vehicle logs show the driver’s hands weren’t detected on the wheel for six seconds before the collision, and he took no action despite having five seconds and about 500 feet of unobstructed view of a concrete highway divider.
and he took no action despite having five seconds and about 500 feet of unobstructed view of a concrete highway divider.
I suspect he may have been distracted or otherwise not paying direct attention, thinking "the car will drive itself, I know I need to watch it but I'll probably be fine if I don't watch it all the time"... and just happened to pick a very wrong time to not pay attention.
...and even when he did realise he was going to crash, there might've been a very strong https://en.wikipedia.org/wiki/Buridan%27s_ass effect which prevented him from quickly deciding which way to turn the wheel.
But they do not take responsibility themselves. It's apparent and in my opinion shameful.
If Tesla's goal of this is to just share the facts, why didn't they state what AutoPilot did? In addition to statements like "...the driver’s hands were not detected on the wheel for six seconds prior to the collision.", wouldn't it make sense to have a statement like "AutoPilot seems to have made a grave error here, and steered the car into the barrier."
They never admit that their car didn't perform properly. This reads like a carefully crafted, legal approved statement - not the more human statements that we see many times after tragedies. This absolutely lowers my opinion of Tesla by a decent amount.
"We feel very bad about what happened... and want to take responsibility and do what's right" - A quote from the Walmart CEO after a Walmart truck accident injured Tracy Morgan and killed another man. I wish Tesla could stand up and say the same in this case (if the evidence points that way, which it appears to).
It was a median in the middle of the highway. The car probably continued in a straight line instead of "steering into" a barrier. https://imgur.com/a/iMY1x
Why the car was not able to detect and avoid the collision in such a simple situation (from a human POV)? We have seen Tesla's autopilot performing very well in a much more complex scenario than this. Ref:- https://www.youtube.com/watch?v=FadR7ETT_1k
Note: Both left lanes are HOV lanes. #1 leads onto Hwy 85, #2 onto Hwy 101. Do they say which lane he was in before entering the median?
Just curious why autopilot would be in lane #1 if he intended to stay on Hwy 101. Or vice versa. Maybe it couldn't make a safe lane change sooner? (Pure speculation)
Fair. Could be rewritten as "AutoPilot seems to have made a grave error here, and drove the car into the barrier. AutoPilot should have either avoided the barrier or disengaged."
Not excusing tesla but how on earth can there be a piece of flat concrete instead of a v shaped bendable deflector ? Was there one but had to be replaced ?
There is a crushable barrier there. It was broken that day, because someone had recently crashed into it. Drove past there yesterday, it has been replaced.
It is pre-mature to say this. Why would they say "AutoPilot seems to have made a grave error here, and steered the car into the barrier." if it is possibly not their fault ?
If indeed it was auto-pilot's fault, they will likely put the statement out later.
What you are asking them to do is legal suicide. If they were to claim this was their fault, they would be immediately slapped with multi-million dollar lawsuits from everybody who ever had a problem with auto pilot. It would end the company.
Really? I don't think so. It's not like a "confession".
Can't people sue them even without their admission of any responsibility whatsoever? I think so.
And that they simply say: "We are sorry, we take full responsibility." does not open them up to lawsuits -- the courts still have to hear the arguments to back up "actual legal liability" and then adjudicate / elaborate a verdict / judgement, right?
I notice they don't mention that the driver complained that the autopilot had trouble with this specific spot on the highway a few times before. https://news.ycombinator.com/item?id=16719736
This is a very deceptively crafted sentence. if i drove on AP for 15 minutes and i was alerted to put my hands on the wheel 2 minutes in.
It would fit tesla's specially crafted statement.
It makes it look like user was warned and didn't respond by putting their hands on the wheel but its completely false. The AP drove him straight into the barrier with no warning.
But this deceptive statement worked as Tesla carefully planned. look at Jalopink title "Tesla Says Autopilot Was On Before Fatal Model X Crash, But That Driver Didn’t Abide Warnings"
PLEASE stop comparing crash statistics for a car with MSRP $80-150k to the general population. Sample bias (among other basic problems) is such a basic consideration that Tesla ignoring it makes me sick. This company will do/say anything to manipulate public perception. People on HN should at least attempt to see through this rather than buying it because Tesla gives them the warm fuzzies.
At 9:30 am, the left lane (lane #2, that stays on 101) is usually packed. He may not have been able to merge in from the #1 lane (HOV-to-Hwy85-flyover lane) into that left lane. Especially if there was a big speed differential.
Although 150 meters is plenty to stop in the median.
(Do we know which lane he was in before hitting the divider?)
In keeping with the "Autopilot" terminology, this was "Controlled Flight info Terrain".
Tesla demonstrates their usual abuse of statistics. 1.25 fatalities per 100 million miles is the average across all types of terrain, conditions, and cars.
The death rate in rural roads is much higher than in urban areas. The death rate on highways is much lower than average. The death rate with new cars is much lower than average.
The autopilot will not engage in the most dangerous conditions. This alone will improve the fatality rate, even if the driver does all the work.
Tesla cars are modern luxury cars. They appear to have done a great job building a well constructed, safe, car. This does not mean their autopilot is not dangerous.
I completely agree; Tesla’s marketing modus operandi is to fudge statistics. In my view, they are consistently on the wrong side of the marketing puffery | misrepresentation borderline.
I do take issue, however, with your claim that Teslas are “well constructed” - they are not. They are poorly designed, poorly constructed vehicles that feel and look cheap. Test one side by side with, say, a BMW 3-Series or a Volkswagen Golf and the difference in production quality will be palpably obvious.
I saw the argument that a traditional motor’s noise level hides rattling sounds and similar subtle car defects, which a fully-electric vehicle’s silence does not cover.
> This does not mean their autopilot is not dangerous.
It also doesn't mean that it is dangerous. That bag of doritos that you ate, it might also be dangerous. Let's use facts and not vague worrisome complaints. So if you want to talk about the higher danger of rural roads, please give a number, and then give a number for tesla, or estimate one. Don't just say "urgh".
I mean, if Tesla is comparing their luxury car's death rates to motorcycles, which have 10x-50x the fatality rate of passenger cars, then, yes, it's dangerous.
I wonder if it’s possible to determine whether those cars have zero death rates in part because someone who would buy that model is less likely to get in a fatal accident in the first place—whether because of who they are, or how they were advertised, or whatever. I imagine a minivan owner as a very cautious driver…then again, I have been in the passenger seat when a friend was doing drifts in his mom’s minivan, so your mileage may vary.
Fatalities are rare events. They chose cars with 100,000 registered vehicles or more. The categories these vehicles belong to experienced around 30 deaths per million cars, so cars that have 100,000 registrations would expect 3 deaths. You could easily get a few models with 0 deaths just as a matter of chance.
I don't think these results necessarily show these cars are inherently safer than other cars in the same category.
Outside of the Tesla marketing department, conditional probability has been known for 300 years. Comparing the average fatality rate to the fatality rate of a modern car with Autopilot is an egregious abuse of statistics. How can anyone defend this practice ?!
Just as an example, click on California (it is mostly consistent all over) on the map [1]. California has a lower average fatality rate of 1.01 per 100 million miles.
Rural roads have 2.62 fatalities per 100 million miles.
Urban roads have 0.70 fatalities per 100 million miles.
The fatality rate is much lower on highways, according to Wikipedia [2], freeways have 3.38 fatalities per billion km (0.54 per 100 million mile, if I managed the conversion).
Note: Tesla is comparing the fatality rate of autopilot-equipped cars with the average accident rate. They are NOT only counting miles where autopilot is used, they are counting all miles driven by their cars. It could be that Tesla drivers are inherently safer than the general population, or that the cars themselves are much safer. What they are not doing (at least with the data) is making a statistical claim about autopilot safety.
Exactly. Tesla drivers will be older, more affluent, and I bet, generally safer drivers. I'm sure an insurance company could easily debunk those statistics.
That's a very good point. But it raises a new question, why didn't Tesla compare the safety record of Teslas with Autopilot against Teslas without Autopilot.
I don’t know. It could be because the comparison isn’t favorable, or it could be because it’s very hard (as your parent post states) to generate a valid comparison that controls for driver age and skill, driving conditions, and vehicle safety.
What I do know is that Tesla have previously published a comparison of the safety of their cars before and after the autopilot feature was made available, and there is a statistically significant improvement of nearly 60%. This study should factor out most of those caveats, since it’s the same cars, same drivers and same roads before and after the feature release. http://bgr.com/2017/01/19/tesla-autopilot-crash-safety-stati...
They have done so. As mentioned in their previous blog post, Tesla's crash rate was reduced by 40% after introduction of Autopilot based on data reviewed by NHTSA.
You realize killing people less than average is not dangerous. It is actually safer. Without understanding this basic logic, than I am not sure what else to say.
I have a lot of respect for Tesla, but this incident convinced me not to buy one any time soon. I have children with car seats. The fires that break out with battery right under the middle makes it impossible to save someone in the driver seat, let alone in the back.
More importantly, self driving cars are not the same - Waymo has lidar, Tesla and Uber don’t (correct me if wrong). If a system is clearly struggling with the detection of still objects in the middle of a highway, on a sunny day, it cannot be trusted. If I ever bought a Tesla, I wouldn’t use autopilot unless their hardware proves reliable.
If you look at the pictures, I don't think much cutting was needed, and certainly any that might be would be nowhere near the batteries located beneath the car; the front end basically disintegrated completely from the impact.
Not quite. The battery pack actually has an aerogel like insulating layer above it in between the cabin and the pack to allow people time to exit, and ports the flames to the front and rear of the vehicle, rather than the sides for this exact reason. It will obviously eventually consume the whole vehicle, because lithium fires burn so hot, but they've done a lot to ameliorate this concern. I thought the driver in this case was removed _before_ the fire started.
"The driver had about five seconds and 150 meters of unobstructed view of the concrete divider with the crushed crash attenuator, but the vehicle logs show that no action was taken."
This seems to be trying to suggest that the driver had 6 seconds of clear driving-toward-death time to correct for the car's actions without explicitly making such a ridiculous statement (while also throwing in the crushed crash attenuator). If the car makes a quick change in direction due to an autopilot error, a driver at speed would have very little time to make an effective correction.
Depending on the system behavior, it could be akin to having a passenger reach over and yank the wheel. I'd honestly rather Tesla just said "ongoing investigation" instead of being so transparently evasive.
They also said in their first post they would respect the family by not updating until the investigation was over. The investigation isn't over but here they are trying to manage their image with confusing and misleading partial information.
Yup this also mean that Tesla's radar and camera systems had 5 seconds to realize it was driving straight into an unmoving concrete barrier and did nothing...
What's really odd is, in my Subaru which has eyesight, if I drove straight towards that barrier the breaks would engage. Eyesight uses two visual cameras, about two feet apart, and uses the parallax between them to judge depth/distance (in the same way human eyesight judges depth).
So even if Tesla's autopilot steered the vehicle towards that divider, shouldn't the auto-braking system have engaged to avert the accident? Tesla vehicles also have visual image cameras as well as radar based ones, so the information density is even higher than Subaru's system.
I guess what I am getting at is: Is auto-braking disabled while autopilot is on? Wouldn't leaving auto-braking enabled (particularly using the visual cameras) offer a second layer of safety is autopilot made an error?
I raised the same issue when a Tesla with autopilot on drove straight into the side of a truck and the driver was decapitated. The discussion was all about "well radar couldn't distinguish it from road signs!" while ignoring that a Tesla has visual (optical) cameras front and center.
In the case of the truck, it was established that neither the driver nor the cameras would have identified the side of the trailer against the sky, and it's still ultimately the driver's responsibility to be alert and avoid such crashes. See the report here: https://static.nhtsa.gov/odi/inv/2016/INCLA-PE16007-7876.PDF
From the report “NHTSA’s crash reconstruction indicates that the tractor trailer should have been visible to the Tesla driver for at least seven seconds prior to impact.”
Unless someone is operating trucks with adaptive optical camouflage, I think that the driver would be expected to note the presence of a big rig. Binocular cameras and radar may be expected to do the same, too.
(Note: Tesla models, to the best of my knowledge, do not have binocular cameras)
The auto-braking in Subarus with eyesight can engage in two circumstances that I'm aware of.
1: The car is travelling at 50 km/hr or less, and perceives an oncoming obstacle.
2: The car is using adaptive cruise control, and perceives that it is getting too close to the vehicle ahead.
I think there are some circumstances where the designers decided that auto-braking would actually be worse. For instance, at highway speed, it's usually not possible to come to a stop quickly enough to avoid a stationary object that has somehow suddenly come into your path. Say, a deer. It's best to take your feet off both pedals and swerve around the obstacle. This reduces speed while also allowing maneuvering, something that braking works against. It also gives drivers behind you more time to react.
If the Tesla system works similarly I would expect similar design decisions. However, I would expect autopilot on a Tesla to function similarly to adaptive cruise control, which will jam on the brakes when it thinks it's going to hit something. This has actually engaged erroneously on two occasions for me, on windy roads, so I only use adaptive cruise control in straight-arrow highway situations now.
I've noticed that in certain weather conditions the cameras will stop being able to get good enough data to make decisions. The car disables the system and notifies me with a warning indicator.
tl;dr - auto-breaking based on parallax cameras can't be deployed in every situation, and sometimes the cameras can't get a clear shot.
The biggest failure here is Tesla's marketing department. Calling it autopilot implies full level 5 autonomous driving in the average person's mind. Very few will read the fine print.
In reality, the current state of it is level 2 or so -- like an enhanced cruise control. You absolutely can't take your eyes off the road still which is what apparently happened here.
Yeah, when driving an emergency vehicle, in my case an ambulance and fire engine, you take a course that used to be called Emergency Vehicle Accident Prevention.
It's now called Emergency Vehicle -Incident- Prevention.
The thing I don't understand is the driver has complained about AP not working on this same location multiple times in the past yet still kept the hands off the wheels for more than 5 seconds?
446 comments
[ 4.5 ms ] story [ 290 ms ] threadShouldn't you have more audible warnings than visual if you are aware that not complying will lead to an accident?
If the autopilot does not think it is capable of driving safely, it should pull over to the shoulder of the road, not beep and flash.
I am not sure if Tesla's current system, which is still just cruise control with steering, should ever be changing lanes unprompted. I don't have a Tesla but I would not want my car's cruise control to do that right now. I do think that pulling over in safe conditions is something that a self-driving system needs to be able to do at some point on the path to full autonomy.
I'm guessing that if I buy a Tesla with autopilot, I'm not going to find any insurers who will charge me discounted rates that reflect autopilot being 10x safer.
making it look like user was warned and didn't respond by putting their hands on the wheel but like i previously said, its completely false.
These kinds of unequivocal statements are just begging for a lawsuit, imo. They simply don't know that this is true.
> In the US, there is one automotive fatality every 86 million miles across all vehicles from all manufacturers. For Tesla, there is one fatality, including known pedestrian fatalities, every 320 million miles in vehicles equipped with Autopilot hardware. If you are driving a Tesla equipped with Autopilot hardware, you are 3.7 times less likely to be involved in a fatal accident.
Let's put aside the consideration of how the average Tesla driver differs compared to the average American driver, or how the average mile driven by Autopilot differs compared to the average mile driven by all human drivers; I'm a little bit bothered by the language here:
> If you are driving a Tesla equipped with Autopilot hardware...
> every 320 million miles in vehicles equipped with Autopilot hardware...
Why is the language "equipped with Autopilot hardware" used? Tesla isn't able to give an estimation of when Autopilot was equipped and engaged?
However, these questions don't lead anywhere interesting. Model S and X cars were driven before and after Autopilot was activated (no hardware change), and the crash rate was reduced after activation. This is the "40% reduction" mentioned by Tesla in the article you pulled the quotes from. You can read a summary here: https://www.theverge.com/2017/1/19/14326258/teslas-crash-rat...
The Insurance Institute for Highway Safety says automatic braking — which can bring the car to a stop if it senses an imminent collision — can reduce rear-end crashes by 40 percent.
That safety advantage can be wiped out in a single accident. Tesla has been lucky that all 3 have been occupied by a single person.
I really think Tesla is jumping the gun with autopilot.
The rest may be true, may be germane, or may not be. The only part that matters to me until the results of the NTSB investigation is made public, is that the autopilot did this.
When the NTSB is investigating an aircraft or train accident we don't see the vehicle manufacturer rushing out a "well we told him to take control" statement and a dump of telemetry. At most they state "we are of course working with the authorities to establish the cause and any corrective actions" until the investigation is completed.
Tesla needs to mature from a blame-the-user software-industry mentality to something more befitting their responsibilities.
What I think they fail to address, especially in this case, is that the autopilot did something a human driver who was paying attention would never do. Autopilot does a great job of saving people from things even a wary driver would miss, much less a negligent one, but the fatal accidents that occur in statistics are not from people who are fully watching the road missing the fact that there's a concrete barrier with yellow safety markings directly in their path, and hitting it head on for no good reason (e.g. evasive action because of another driver, etc, oh a stupid last-minute "oh crap that's my exit").
I want autopilot to succeed, and I want Tesla (and Musk) to succeed, and for the sake of their public image they have to realize that this isn't an average accident statistic, a lapse in attention or evasive maneuvering. It's a car that seemingly plowed right into a concrete barrier while still under complete control. That's not a mistake a healthy human will make.
Then how did the crushed barrier get crushed before the Tesla hit it? Clearly, the stretch of road is unsafe enough to trick humans drivers (and, clearly, Tesla should improve).
The most likely scenarios I can think of would be not paying attention and drifting into the barrier, attempting to avoid a car merging into the lane (and not paying enough attention), or being struck by another car and being forced into the barrier.
https://www.google.com/maps/@44.436552,25.9597006,18z
https://techcrunch.com/wp-content/uploads/2018/03/screen-sho...
Reminds me of a scene in "La grande vadrouille" where a motorcyclist is killed on a mountain road because he was following the dotted line and the painter had gone to the side of the road to take a break.
Also, that's an immense pothole!
... well you certainly aren’t from southeast Michigan, that’s for sure.
https://techcrunch.com/wp-content/uploads/2018/03/screen-sho...
I think 3d mapping the surrounding world correctly using multiple cameras (what humans are doing) is more generalizable.
There's a spot just like this in Houston on 610 East. HUUUUUUGE flyover as the HOV lane spends about a half mile merging into the left lane of 610 (nice fat fast freeway, they need the runway). A guy had thought the flyway would be a safe place to park his car and wait for a tow. Safe enough that he was sitting on the hood of his car. Sure as shit the Challenger in front of me got confused, cut into the flyaway thinking it was a lane, hit the disabled car, and sent the guy flipping up into the air a good 10, 15 fit. Fucked up accident.
Anyway point is after the accident I was talking to the driver of the Challenger, trying to calm him down (kept saying "holy shit I fucking killed that guy!") and other that shock he was sober as a duck. He got bloodtested and everything, clean. In court he said he just got confused and thought it was a lane.
LONG STORY SHORT human error mang, human error. Still not sure why a car was able to do it.
They have those same white lines leading into intersections and people cross that shit all the time. People speed.
The vehicle should have at the very least stopped short of the road obstruction. But we don't yet know what led up to being in that lane, I was hoping the article in the OP would have taken us through it. Instead I'm getting the feeling this was a catastrophic software failure. Without more transparency on the issue, despite what Tesla may prefer I don't think I have any choice but to feel uncomfortable.
The divider is from the fast lane no less.
Though of course that doesn't explain why it didn't recognize the barrier as something it should avoid at all costs. Unless perhaps something to the left confused it and made it think there was an obstacle there, too, thus causing it to think it was going to crash regardless. If so, maybe it did try to brake at the last second?
A piece of software that mistakes that wedge for a lane in broad daylight has no business being deployed on the road, in my opinion.
The worst such crash I've ever seen was barely 2 miles from there, on northbound 85 near Fremont Ave. There is a soundwall that comes to a connection point where a wall segment is directly perpendicular to the freeway. For some reason, the guardrail had a gap there almost exactly the width of a vehicle.
A few years ago, a vehicle veered off the right shoulder and perfectly threaded the gap, into the wall at full speed. It was compressed to maybe 5 feet long.
[1] http://abc7news.com/automotive/exclusive-i-team-investigates...
Either:
A) The autopilot could not see a concrete barrier in its path. B) More likely, and as the story reports, it WAS aware of the danger but didn't do anything.
Either case is at least worth discussing, no?
The only reason a human drives into that barrier is if they swerve to avoid something or are otherwise not properly paying attention (texting, etc.).
There's no good reason for the autopilot to have hit that barrier.
The mental gymnastics going on in these comments attempting to absolve Tesla of any responsibility are truly next-level.
Again, not absolving Tesla, just being realistic about the capabilities of human drivers.
If one system is steering the vehicle into things that the other system would, in most conditions, reliably avoid, it bears some discussion.
Tesla's statement is not clear on this at all.
If that was the case, they would have said so. Their vagueness here is telling.
true. here’s the location (post #62)
https://teslamotorsclub.com/tmc/threads/model-x-crash-on-us-...
but the average person in the US anyway is not a good driver. i very often see people get aurprised at the lane ending at that exact spot. they should put a rumble strip leading up to it.
So, my question is, is it fair to compare AP to a human driver paying attention or is it more fair to compare it to an average driver.
I mean that just for the sake of argument. In this specific case, NTSB or someone should ban AP. It is so obvious what’s about to happen there, an AP should do what a driver should do which is take the ramp whether it’s a wrong turn or not. So many a-holes try to squeeze in last second (oops) when they should just take the damn exit or miss the damn exit as the case may be. What AP did here is what a poor and panicked driver would do and that’s just not acceptable.
Humans do exceedingly stupid things all the time because they stop paying attention, even momentarily (or subconsciously).
We put big lights on the back of cars that light up when they brake. And yet, despite a driver looking direct at a huge object with two lights, that rapidly grows larger right in front of them, does not always prevent a collision. Or even a chain of collisions. I don't get on the road much, and yet even I've seen a ton of accidents that are baffling and can only have been the result of a driver not paying attention for a bit.
I'm reminded of how in quite a few places removing signs and lights actually improved safety because it forced drivers to stay aware instead of 'driving on autopilot', so to speak.
I think that's the real issue here. The more we outsource our attention to a machine, the more important it is that said machine does MUCH better than we humans do. Especially if a mistake can be deadly.
But I wouldn't be surprised if, indeed, technically this accident could've happened just as easily by a non-autopilot car where the driver had a little 'micro-sleep', got distracted by something in his field of view, mistakenly thought he was on a lane and didn't notice the (let's be fair) ridiculously bad markers that were the only way to tell that part of the 'gray' stuff in front of him was in fact a wall of concrete.
I mean, just look at the image: https://imgur.com/a/iMY1x . Half of what makes the barrier stand out is the shadow!
All that said I might sound more argumentative than I am. I do agree with most of your comment.
I'm sure that out of the 1.25 million annual automobile-related deaths, plenty of drivers were paying attention and still did stupid things similar to this accident.
"this isn't an average accident statistic, a lapse in attention or evasive maneuvering. "
"That's not a mistake a healthy human will make."
Almost every driver thinks they're significantly better than average. Few are.
If a human driver's lapse of attention causes a similar crash, how much less of a tragedy is it just because we can less ambiguously blame the victim?
My opinion is that the statistics compare just fine.
Actually, about half of drivers are better than average.
This statement is meaningless circular logic. You can handwave away any incident with a human driver by saying he wasn't "healthy".
A human pilot has intentionally driven an airplane full of passengers into the ground with full control because he wanted to kill himself. The airline believed he was "healthy".
So, I thought it wasn't all that clever in the first place to try and marry the risks of getting electric cars into the market with the risks of telling the extremely wealthy that they didn't need to hold the damn steering wheel.
Would've loved to know the exact moments we are talking about here. Is it 5 seconds or 5 minutes?
> The driver had received several visual and one audible hands-on warning earlier in the drive and the driver’s hands were not detected on the wheel for six seconds prior to the collision.
What does this mean? Did Tesla want to give over control to driver? Or just normal no hands?
> The driver had about five seconds and 150 meters of unobstructed view of the concrete divider with the crushed crash attenuator, but the vehicle logs show that no action was taken.
Again its hard to know if Tesla wanted to give up control or not. I hope they aren't just saying that because the driver's hands were off the steering wheel, the crash occurred and is his fault for not obeying the agreement.
That's exactly Tesla's official stance - if you crash you're SOL and at fault, Autopilot or Auto Emergency Braking be damned.
https://www.cnbc.com/2018/01/31/apples-steve-wozniak-doesnt-... "Man you have got to be ready — it makes mistakes, it loses track of the lane lines. You have to be on your toes all the time," says Wozniak. "All Tesla did is say, 'It is beta so we are not responsible. It doesn't necessarily work, so you have to be in control.'
"Well you that is kinda a cheap way out of it."
In this case, it would be more like "drive in a straight line and... collide to a stop."
Perhaps braking to a stop at a reasonable rate would be the right thing to do, given that it surely should've detected it was going to hit something?
I need to make a post, as a long time Tesla owner/very frequent autopilot user, because I think I know exactly what happened. Unfortunately I need to go to bed, already up later that I should be for tomorrow. A lot of information and comments I’ve seen about how it works and how autopilot is dangerous etc, are quite wrong. I think it’d be important for people to understand how it works from people who use it and get their opinions first before making statements about AP in general.
For the record, I love using AP and will continue to do so. But it is no silver bullet, and 99.5% of Tesla owners know this and how to correctly use it.
This concrete barrier is also quite a distance past where this left exit lane starts to separate from the rest of US-101, so there are really only two ways you'd risk hitting it. First, and probably most common, is people who change their minds last-second about taking that exit. Second, would be a serious AP screw-up (like what might have happened here).
FWIW, I've never had AP engaged when driving through this spot, and almost instinctively disable it when taking a highway off-ramp/change-over.
And on 'trust' Tesla have a history of being open and upfront. Organisations tend to repeat a culture. So for Uber I would closely question their statements given an ongoing history of unethical behaviour across many levels. It seems likely they will lie or show limited truth to reach an agenda. Tesla's history seems open and honest.
Or the fact that they will remotely downgrade your car (to an older software version) and remove features of it (disable ethernet ports) if they find you "snooping" and discover references to new models.
Thats the kind of openness and honesty you laud them for?
It would fit tesla's specially crafted statement.
It makes it look like user was warned and didn't respond by putting their hands on the wheel but its completely false. The AP drove him straight into the barrier with no warning.
>very deceptively crafted sentence. >The AP drove him straight into the barrier with no warning.
Exactly - anyone who owns autopilot knows this.
The atlantic had a relevant piece on the morality there this morning https://www.theatlantic.com/technology/archive/2018/03/got-9...
And that's assuming they can identify that autopilot is statistically safer. If autopilot isn't used in the same conditions as all human driving, the rates aren't apples to apples comparisons, and the "you are 3.7 times less likely to be involved in a fatal accident" figure they imply is misleading.
Is that true? I couldn't find statistics that grouped by income or wealth.
"The driver had received several visual and one audible hands-on warning earlier in the drive ... and the driver’s hands were not detected on the wheel for six seconds prior to the collision."
https://www.tesla.com/blog/update-last-week’s-accident
Regardless of the criticism of Tesla here, surely the driver wasn't watching the road, or otherwise this wouldn't have happened.
Seems like an oxymoron that's getting drivers into danger.
>Tesla said its vehicle logs show the driver’s hands weren’t detected on the wheel for six seconds before the collision, and he took no action despite having five seconds and about 500 feet of unobstructed view of a concrete highway divider.
And it's not the first time.
I suspect he may have been distracted or otherwise not paying direct attention, thinking "the car will drive itself, I know I need to watch it but I'll probably be fine if I don't watch it all the time"... and just happened to pick a very wrong time to not pay attention.
...and even when he did realise he was going to crash, there might've been a very strong https://en.wikipedia.org/wiki/Buridan%27s_ass effect which prevented him from quickly deciding which way to turn the wheel.
-The driver
-The crushed barrier
-Statistics
But they do not take responsibility themselves. It's apparent and in my opinion shameful.
If Tesla's goal of this is to just share the facts, why didn't they state what AutoPilot did? In addition to statements like "...the driver’s hands were not detected on the wheel for six seconds prior to the collision.", wouldn't it make sense to have a statement like "AutoPilot seems to have made a grave error here, and steered the car into the barrier."
They never admit that their car didn't perform properly. This reads like a carefully crafted, legal approved statement - not the more human statements that we see many times after tragedies. This absolutely lowers my opinion of Tesla by a decent amount.
"We feel very bad about what happened... and want to take responsibility and do what's right" - A quote from the Walmart CEO after a Walmart truck accident injured Tracy Morgan and killed another man. I wish Tesla could stand up and say the same in this case (if the evidence points that way, which it appears to).
https://www.cbsnews.com/video/walmart-ceo-on-tracy-morgan-ca...
This autopilot incident is similar to the one with the tractor-trailer[1] in that regard.
[1] https://www.theguardian.com/technology/2016/jun/30/tesla-aut...
Or the one with the fire truck: https://www.theverge.com/2018/1/23/16923800/tesla-firetruck-...
Just curious why autopilot would be in lane #1 if he intended to stay on Hwy 101. Or vice versa. Maybe it couldn't make a safe lane change sooner? (Pure speculation)
But maybe this is just a variant on https://www.wired.com/story/tesla-autopilot-why-crash-radar/ That is, maybe Mr. Huang was following another vehicle, which merged out of the way, and his Tesla ignored the stationary object.
If that's so, Tesla ought to emphasize the failure mode.
If indeed it was auto-pilot's fault, they will likely put the statement out later.
Can't people sue them even without their admission of any responsibility whatsoever? I think so.
And that they simply say: "We are sorry, we take full responsibility." does not open them up to lawsuits -- the courts still have to hear the arguments to back up "actual legal liability" and then adjudicate / elaborate a verdict / judgement, right?
Comments from lawyers, please. Very interesting.
This is a very deceptively crafted sentence. if i drove on AP for 15 minutes and i was alerted to put my hands on the wheel 2 minutes in.
It would fit tesla's specially crafted statement.
It makes it look like user was warned and didn't respond by putting their hands on the wheel but its completely false. The AP drove him straight into the barrier with no warning.
But this deceptive statement worked as Tesla carefully planned. look at Jalopink title "Tesla Says Autopilot Was On Before Fatal Model X Crash, But That Driver Didn’t Abide Warnings"
> No one knows about the accidents that didn’t happen, only the ones that did.
The same applies to jobs, and yet Obama’s “jobs saved” were cited as major accomplishment of QA. lol...
https://www.technologyreview.com/s/601849/teslas-dubious-cla...
At 9:30 am, the left lane (lane #2, that stays on 101) is usually packed. He may not have been able to merge in from the #1 lane (HOV-to-Hwy85-flyover lane) into that left lane. Especially if there was a big speed differential.
Although 150 meters is plenty to stop in the median.
(Do we know which lane he was in before hitting the divider?)
Tesla demonstrates their usual abuse of statistics. 1.25 fatalities per 100 million miles is the average across all types of terrain, conditions, and cars.
The death rate in rural roads is much higher than in urban areas. The death rate on highways is much lower than average. The death rate with new cars is much lower than average.
The autopilot will not engage in the most dangerous conditions. This alone will improve the fatality rate, even if the driver does all the work.
Tesla cars are modern luxury cars. They appear to have done a great job building a well constructed, safe, car. This does not mean their autopilot is not dangerous.
I do take issue, however, with your claim that Teslas are “well constructed” - they are not. They are poorly designed, poorly constructed vehicles that feel and look cheap. Test one side by side with, say, a BMW 3-Series or a Volkswagen Golf and the difference in production quality will be palpably obvious.
(I haven’t seen scientific comparisons, though.)
It also doesn't mean that it is dangerous. That bag of doritos that you ate, it might also be dangerous. Let's use facts and not vague worrisome complaints. So if you want to talk about the higher danger of rural roads, please give a number, and then give a number for tesla, or estimate one. Don't just say "urgh".
Remember, there are several very normal cars models that have had ZERO deaths: https://www.nbcnews.com/business/autos/record-9-models-have-...
These autopilot cars are death traps.
Fatalities are rare events. They chose cars with 100,000 registered vehicles or more. The categories these vehicles belong to experienced around 30 deaths per million cars, so cars that have 100,000 registrations would expect 3 deaths. You could easily get a few models with 0 deaths just as a matter of chance.
I don't think these results necessarily show these cars are inherently safer than other cars in the same category.
Just as an example, click on California (it is mostly consistent all over) on the map [1]. California has a lower average fatality rate of 1.01 per 100 million miles.
Rural roads have 2.62 fatalities per 100 million miles. Urban roads have 0.70 fatalities per 100 million miles.
The fatality rate is much lower on highways, according to Wikipedia [2], freeways have 3.38 fatalities per billion km (0.54 per 100 million mile, if I managed the conversion).
[1] https://cdan.nhtsa.gov/stsi.htm
[2] https://en.wikipedia.org/wiki/Transportation_safety_in_the_U...
What I do know is that Tesla have previously published a comparison of the safety of their cars before and after the autopilot feature was made available, and there is a statistically significant improvement of nearly 60%. This study should factor out most of those caveats, since it’s the same cars, same drivers and same roads before and after the feature release. http://bgr.com/2017/01/19/tesla-autopilot-crash-safety-stati...
Thanks, I never knew this term before. Seems like a dark humour gem - equally funny and terrifying.
More importantly, self driving cars are not the same - Waymo has lidar, Tesla and Uber don’t (correct me if wrong). If a system is clearly struggling with the detection of still objects in the middle of a highway, on a sunny day, it cannot be trusted. If I ever bought a Tesla, I wouldn’t use autopilot unless their hardware proves reliable.
This seems to be trying to suggest that the driver had 6 seconds of clear driving-toward-death time to correct for the car's actions without explicitly making such a ridiculous statement (while also throwing in the crushed crash attenuator). If the car makes a quick change in direction due to an autopilot error, a driver at speed would have very little time to make an effective correction.
Depending on the system behavior, it could be akin to having a passenger reach over and yank the wheel. I'd honestly rather Tesla just said "ongoing investigation" instead of being so transparently evasive.
So even if Tesla's autopilot steered the vehicle towards that divider, shouldn't the auto-braking system have engaged to avert the accident? Tesla vehicles also have visual image cameras as well as radar based ones, so the information density is even higher than Subaru's system.
I guess what I am getting at is: Is auto-braking disabled while autopilot is on? Wouldn't leaving auto-braking enabled (particularly using the visual cameras) offer a second layer of safety is autopilot made an error?
I raised the same issue when a Tesla with autopilot on drove straight into the side of a truck and the driver was decapitated. The discussion was all about "well radar couldn't distinguish it from road signs!" while ignoring that a Tesla has visual (optical) cameras front and center.
Unless someone is operating trucks with adaptive optical camouflage, I think that the driver would be expected to note the presence of a big rig. Binocular cameras and radar may be expected to do the same, too.
(Note: Tesla models, to the best of my knowledge, do not have binocular cameras)
The auto-braking in Subarus with eyesight can engage in two circumstances that I'm aware of.
1: The car is travelling at 50 km/hr or less, and perceives an oncoming obstacle.
2: The car is using adaptive cruise control, and perceives that it is getting too close to the vehicle ahead.
I think there are some circumstances where the designers decided that auto-braking would actually be worse. For instance, at highway speed, it's usually not possible to come to a stop quickly enough to avoid a stationary object that has somehow suddenly come into your path. Say, a deer. It's best to take your feet off both pedals and swerve around the obstacle. This reduces speed while also allowing maneuvering, something that braking works against. It also gives drivers behind you more time to react.
If the Tesla system works similarly I would expect similar design decisions. However, I would expect autopilot on a Tesla to function similarly to adaptive cruise control, which will jam on the brakes when it thinks it's going to hit something. This has actually engaged erroneously on two occasions for me, on windy roads, so I only use adaptive cruise control in straight-arrow highway situations now.
I've noticed that in certain weather conditions the cameras will stop being able to get good enough data to make decisions. The car disables the system and notifies me with a warning indicator.
tl;dr - auto-breaking based on parallax cameras can't be deployed in every situation, and sometimes the cameras can't get a clear shot.
In reality, the current state of it is level 2 or so -- like an enhanced cruise control. You absolutely can't take your eyes off the road still which is what apparently happened here.
I had intended to buy TSLA after their next awful earnings call and hold it long, now I'm not sure I want anything to do with this business.
http://www.roadpeace.org/take-action/crash-not-accident/
https://www.crashnotaccident.com
It's now called Emergency Vehicle -Incident- Prevention.