Slashdot Mirror
DVD Player Found In Tesla Autopilot Crash, Says Florida Officials (reuters.com)
An anonymous reader quotes a report from Reuters: A digital video disc player was found in the Tesla car that was on autopilot when its driver was killed in a collision with a truck in May, Florida Highway Patrol officials said on Friday. "There was a portable DVD player in the vehicle," said Sergeant Kim Montes of the FHP in a telephone interview. She said there was no camera found, mounted on the dash or of any kind, in the wreckage. A lawyer for a truck driver involved in the accident with the Tesla told Reuters his investigators had spoken to a witness who said the DVD player was playing a "Harry Potter" video after the accident, but the lawyer was unable to verify that beyond the witness account. Lawyers for the family of the victim, 40-year-old Joshua Brown, released a statement Friday saying the family is cooperating with the investigations "and hopes that information learned from this tragedy will trigger further innovation which enhances the safety of everyone on the roadways." Lawyers for the family of the victim, 40-year-old Joshua Brown, released a statement Friday saying the family is cooperating with the investigations "and hopes that information learned from this tragedy will trigger further innovation which enhances the safety of everyone on the roadways." Tesla said in a statement Friday, "Autopilot is by far the most advanced driver assistance system on the road, but it does not turn a Tesla into an autonomous vehicle and does not allow the driver to abdicate responsibility."
What exactly is the point of it? To lull you into a false sense of comfort and security? I look forward to autonomous vehicles, but if it still requires me to keep my attention on the road and ready to respond, I'd rather just be in control of the vehicle to begin with.
Elon Musk's Terminators have claimed their first victim.
How dare you speak like this about Martian Citizen Zero? He is going to lead the entire human species to Mars! Sure, a few airlocks may open on the way, food dispensers may not work, but think how jealous the people left on Earth will be!
The problem is that if it slightly resembles a full-on AI based driverless system, that's how people are going to treat it no matter how many layweresque warnings you thrust in front of them and no matter how many forms they have to sign telling them it is just fancy lane assist.
It's just human nature: if people aren't actively involved in the driving process, their attention is going to wander. It's how we as humans are wired up. For a long trip, I'm not sure I could stay focused at all times, even though I'd know perfectly well I was risking my life if my attention wandered. If I'm driving, that's one thing, but if the car is doing 99.9% of it, the other 0.1% is going to pose a real serious problem.
If you build "almost an autopilot", that is a recipe for people treating it like what it resembles but isn't.
The car was basically equipped with a stay-in-lane and slow-down-if-you-approach-the-car-in-front-of-you kind of system, which is not an autonomous vehicle, nor can you take your eyes off the road. At best it reacts a bit faster if someone in front of you hits the brakes. Google did a talk on this and said in their tests, as soon as a car seems to be working by itself, drivers stopped paying attention to the road, so half-way-autonomous is a bad idea. People don't want to pay attention and they won't if the car seems to be doing a good enough job.
Only a fully autonomous car will be good enough.
"I have never let my schooling interfere with my education." - Mark Twain
Yes, I know, jealousy of Musk is a big motivation for you people to hate on him, however if we ignore that for a second..
The point here is someone knowingly placed their life in the hands of automation, and paid the price for that. Real world
conditions mean that automation is NEVER perfect, and this is new automation at the cutting edge (sigh) of such things.
The larger issue to me is why the DRIVER did not notice a truck across the road in front of them. Are we to believe that
the software should have spotted it, and yet it was so hard to spot that a driver who was paying attention could not? That
would certainly stretch the bounds of credibility quite far.
It seems quite clear here that the driver was not watching the road ahead - in fact was ignoring it enough to not notice a
whole, large truck trailer unit turn in front of them in clear view. In other words they were, unfortunately for them, doing
something stupid.
But no, people are going to try and blame automation, because otherwise it would be a dead person at fault.. And that is
just not nice, right? However, this is NOT a case where a driver jumped on the brakes and they did not work, or tried
to turn the car and it went straight ahead (at least none of that is being claimed). It is a case where a driver of a car
at speed was not aware of the road directly ahead of them, that makes this border on a darwin here folks..
Porn. Absolutely porn. No 40 year old man is driving down the road in his bad ass Tesla watching Harry Potter. No way, not happening. Porn.
Beware of the Leopard.
Although in this particular case it is unclear whether the driver was actually watching a DVD at the moment of the crash, it is pretty obvious that an assisted driving technology that can handle 95% of the driving situations will make users confident enough to be distracted when operating the vehicle, no matter how many warnings and disclaimers are shown telling users they need to pay attention all the time in case they have to gain control to handle the remaining 5% of the traffic situations. This is clearly explained in this TED talk by the head of Google driverless car program: https://www.youtube.com/watch?... (this particular issue is discussed around 4:10, although the whole video is worth watching). This is why Google approach to self driving cars is to release their product when the system is able to handle 100% of the driving situations and never require the user to take control in contrast to the Tesla approach of releasing a system than can handle most situations and make incremental improvements over time.
It is a case where a driver of a car
at speed was not aware of the road directly ahead of them, that makes this border on a darwin here folks..
Not necessarily. Maybe the rest of the automation had been so good that the driver saw the struck, but believed the car also saw the truck. If you are a passenger in a car, you don't pull the handbrake to avoid an accident when you expect the driver is going to press the foot brake.
That said, he was probably just watching Harry Potter.
Real lawyers write in C++
"Autopilot is by far the most advanced driver assistance system on the road, but it does not turn a Tesla into an autonomous vehicle and does not allow the driver to abdicate responsibility."
Then maybe they should start by stopping to use the misleading name of "autopilot" for this functionality.
The problem is not the Autopilot feature but the way it has been misleadingly and dangerously marketed.
Musk bragged to the press that Autopilot was "almost twice as good as a person," certainly sending the wrong message. His ex-wife posted a YouTube video of her driving while covering her eyes and dancing around while on Autopilot on a crowded highway. All this has encouraged a bunch of other YouTube videos of people behaving foolishly while on Autopilot.
https://www.yahoo.com/news/tesla-mixes-warnings-bravado-hands-free-driving-002343250--finance.html
Even the marketing name "Autopilot" is probably misleading to some people, who might interpret as "the car drives itself without human assistance". It should have been more conservatively called "driver assist" or some such.
In the end their marketing stupidity is probably going to bite them financially. A dashboard warning doesn't excuse it. I say this regretfully as a Tesla stockholder.
but that does not change the fact that they released a fundamentally flawed and extremely dangerous product.
At this stage, that is your opinion and not a fact. Don't purport it as such.
FYI, it's envy, not jealousy. Jealousy would mean you were afraid of losing Elon. Envy means you wish you had all his cool stuff, fun life, and hot ex-wife. I only make this correction because I found out I was saying it wrong for 30 years...
The larger issue to me is why the DRIVER did not notice a truck across the road in front of them.
How can we be sure that the driver did *not* notice the truck?! The fact that he didn't step on the brakes, you say? Maybe he noticed it, but thought "whatever, my Tesla is smart enough to stop if it needs to". And by the time he realized that it's not stopping he just didn't have enough reaction time to lift the foot off the floor and apply the brakes? Every time I see a demo of the smart cruise control, where the car can stop if there's an obstacle, drivers are told to resist stepping on the brakes and trust the car to slow down and stop in time. Or are we supposed to hover our foot over the brake pedal and second-guess the car all the time? What's the point of having an auto-pilot if you can't relax and let the car take care of the brakes?
As much as I like Tesla (my next car was going to be one until this happened), I think they should suffer enough for this, so that every other automaker out there makes damn sure their auto-pilot can handle situations like this. And I'm waiting another 5 years before getting a car with auto-pilot.
The fact is that Tesla states that: ""Autopilot is by far the most advanced driver assistance system on the road, but it does not turn a Tesla into an autonomous vehicle and does not allow the driver to abdicate responsibility."
According to the GP, while taking human psychology into account, this is what makes this a fundamentally flawed and extremely dangerous product. People will watch Harry Potter movies in this car, they will have horrible response times because they don't need to pay attention, they will get into accidents when the 'driver assistent' fails, and Tesla will try to abdicate responsibility each and every time based on contractual terms.
Then that is Darwin award territory.
I do not understand. Why would you think they do in the first place? Perfect SF movie artificial intelligence has not been invented and installed in a car. Are you being serious?
I am a bit surprised about the belief that AIs (or machine learning) will solve all problems given enough data.
What do you think a neural net would have learned to do if trained to use VW's "AdBlue" as efficiently as possible but still to pass the NHTSA conformance test?
Who would you blame then? After all the constraints look reasonable. Would you want to be the engineer sued because he did not predict the neural net might learn something illegal?
Plus, there is obviously a problem with the way Tesla gathers its training data. If Elon Musk promotes a dashcam video taken by the killed driver earlier where the driver admits insufficient attention to the road (the cutting-in vehicle was in front of the driver and clearly visible), people might well take this as encouragement to not pay attention.
But no, people are going to try and blame automation, because otherwise it would be a dead person at fault.. And that is
just not nice, right? However, this is NOT a case where a driver jumped on the brakes and they did not work, or tried
to turn the car and it went straight ahead (at least none of that is being claimed). It is a case where a driver of a car
at speed was not aware of the road directly ahead of them, that makes this border on a darwin here folks..
Sure the person was at fault for paying attention while driving.
But people not paying attention while driving is the obvious outcome of giving a car an "autopilot" that operates on highways.
I stole this Sig
TCAS is only a warning system and that's unilkely to change any time soon.
Airplanes on autopilot will happily fly into other airplanes, mountains, buildings, etc... TCAS and GPWS will give aural and visual warnings, but that's it. If the pilots ignore those warnings, the autopilot will just continue on its path.
By the way, TCAS relies on transponders in other aircraft that broadcast their position and altitude. Cars don't have those.
He was confused as to where platform 9 3/4 was.
Unfortunately the Tesla system did not even provide any warnings, you really can't have a similar warning system as planes because you will typically have a lot of cars in the vicinity. So comparing car to plane autopilot is very applish-orangish. Plane autopilot is actually a lot simpler to accomplish.
Allowing the user to have hands off for 30 seconds is problematic for Telsa. A lot can happen in 30 seconds, its an arbitrary duration. Why not 5 seconds?
imho, it should not be called auto-pilot or autonomous driving because its not truly that yet. Assisted control is more appropriate.
I have to remember that old joke. From Soviet times, when else
GDR-FRG border. GDR subordinate storms into the office of his superior.
Sub: Comrade! The Russians, they're on the moon!
Sup: All of them?
Sub: No... they just sent a capsule up.
Sup: Then why the fuck do you wake me? Just report when they're all gone.
We used to have a Bill of Rights. Now, with the rights gone, all we have left is the bill.
Another thing that probably could have prevented this: raw camera data.
There's no way that a truck was reflecting exactly the same amount of R, G and B (let alone full spectrums) as the sky. But it probably was the same pixel in the image taken by the camera: 255,255,255 (or higher if it's more than 8 bits). The process of converting raw data to linear space tends to truncate both dark and bright pixels; in reality you may have one pixel of a dark object in a shady place indoors that's numerous orders of magnitude lower light intensity than a pixel of a white object in direct sunshine out the window. Camera data should be returned to the autopilot system in raw format - either direct photon counts, or a floating point representation of total activation on the pixel.
The more data about what's in the pixels they can return to it - not just simply more resolution - the better. After all, more resolution may let you make out shapes further away, but the car here didn't even make out the truck close up. They should be collecting a broader spectrum and polarization data. Which may sound very difficult, but really it's not. CCDs are already IR sensitive - it's normally something that they have to try to work around by IR filtration. We already have numerous filters for different wavelengths and polarization filters that can be used in addition to the standard RGB filters - converting resolution to better spectral data. You can be assured that at least on *some* channel, with raw data, the truck would have been visible. And if not, then there's no way on Earth a human could have seen it.
But I think the problem here is much more fundamental then a lack of extra channels. It sounds clear to me that they had the horizon and truck wash out.
Did he just go crazy and fall asleep?
More to the point about spectrums... mid to near IR should show operating vehicle engines, potentially exhaust, etc as hot pixels. And long-wave IR should show people and animals as hot pixels. Both of which sound *incredibly* useful.
That said, I'm not sure where traditional CCDs stop being sensitive... I imagine they don't go all the way down to the long-wave spectrum. They do of course make cooled IR cameras that capture long-wave but they tend to be larger and more expensive. Hmm, let's see how far traditional uncooled CCDs can go... I'm seeing a number of pages putting the range limit at around 150 (or 300?) to 1100nm (human vision is 380 to 750nm, give or take). I wouldn't be surprised if some parts of an engine would glow reasonably well in the 1000+ nm range.... but that's *if* you could see it, though, without something blocking the radiation. I doubt they could see exhaust, at least at the point it leaves the tailpipe. You'd need a special designed, cooled camera if you want to see the lower ranges.
Did he just go crazy and fall asleep?
Regular lens glass isn't great for IR.
Many of the best IR lens materials can't stand humidity.
http://www.edmundoptics.com/re...
This was a software issue. The camera 'saw' the truck, but the edges didn't have high enough contrast.
Fundamentally though, this was an autopilot induced crash. If the driver had continued to pay attention, he could have avoided it. Airplanes had the same problem, driver assists are dangerous if they allow the driver to feel safe when not focused on driving.
John McAfee 'It was like that time I hired that Bangkok prostitute; to do my taxes, while I fucked my accountant'
Again, it depends on what you mean by "IR", which is a very broad spectrum range. Cameras often have to add a special IR filter to block near-IR because the lens doesn't block it on its own. You can see here the transmission spectrums of different types of glasses and plastics. You can see that as a general rule they're good at blocking UV but not IR, at least near-IR (750-1400nm). They tend to block more IR the closer you get to the far-IR spectrum, however.
Images being overexposed will do that to you. And the overexposure of an image isn't a fundamental aspect of CCD hardware, it's a processing artifact.
Example: take this image. Note how the boundary between the car and the sky in this picture is completely lost. It's not like the CCD is receiving the exact same amount of photons from the car and the sky - they're actually going to be very different. But they're both truncated off at maximum brightness when saved into an "image" - and that image is then provided to the autopilot. In severe cases, the autopilot is highly disadvantaged, if not inherently doomed, no matter how good its software is. Human eyes don't have that limitation - we can see bright and dark areas simultaneously and make out details in both.
The CCD is getting the data that's needed. But the autopilot isn't.
Did he just go crazy and fall asleep?
You mean he purchased an automobile?
The airbag blew up in his face?
No. It is a common tactic in most industries to imply human error was the cause immediately after an accident. This quickly placates the general public. When the results of the investigation prove the claim was unfounded, that information doesn't get remotely as much publicity, and nearly everyone has the incident committed to memory as 100% human error.
People driving completely non-autonomous cars have accidents all the time, including ones just like this. The quote from Musk in the first story was that the car's sensors couldn't distinguish a white trailer on a bright day. Obviously that can be difficult for human eyes as well, particularly around sunset.
Slashdot gets worse every day... Pipedot: News for nerds, without the corporate slant
6 seconds