Slashdot Mirror


Elon Musk: Autopilot Feature Was Disabled In Pennsylvania Crash (latimes.com)

An anonymous reader writes: In response to the third reported Autopilot crash, which was the first of three where there were no fatalities, Tesla CEO Elon Musk says that the Model X's Autopilot feature was turned off. He tweeted Thursday afternoon that the onboard vehicle logs show that the semi-autonomous driving feature was turned off in the crash. "Moreover, crash would not have occurred if it was on," he added. The driver of the Model X told police he was using the Autopilot feature, according to the Detroit Free Press. The vehicle flipped over after hitting a freeway guardrail. U.S. auto-safety regulators have been investigating a prior crash that occurred while Tesla's Autopilot mode was activated. Late Thursday afternoon and into early Friday, Musk made some comments on the improvements made to its radar technology used to achieve full driving autonomy. "Working on using existing Tesla radar by itself (decoupled from camera) w temporal smoothing to create a coarse point cloud, like lidar," he tweeted. "Good thing about radar is that, unlike lidar (which is visible wavelength), it can see through rain, snow, fog and dust." Musk has rejected Lidar technology in the past, saying it's unnecessary to achieve full driving autonomy. Consumer Reports is calling on Tesla to "disable hands-free operation until its system can be made safer."

34 of 166 comments (clear)

  1. Hands-free? by Guspaz · · Score: 3, Insightful

    Tesla cars don't support hands-free operation. You're supposed to keep your hands on the steering wheel while using autopilot, and the car will disable auto pilot after a while if you take your hands off the wheel.

    Perhaps they should reduce that timeout to discourage people from taking their hands off the wheel entirely.

    1. Re:Hands-free? by robbak · · Score: 4, Informative

      If it does not detect your hand on the wheel for a certain number of seconds, it alerts to tell you to return your hands to the wheel, and, if you don't, then it slows the car down to a gentle stop.

      --
      Prediction for end of Universe #42: Fencepost error in Quantum_bogosort.cpp
    2. Re: Hands-free? by WindBourne · · Score: 3, Informative

      interestingly, that is exactly what was happening in Penn. the driver had his hands off and it was at the point of turning off stereo and slowing down. Then when driver put his hands on it, he turned it off and then took it off road. IOW, this is another fuckup who wants to blame Tesla and sue.

      --
      I prefer the "u" in honour as it seems to be missing these days.
  2. Re:Beyond a doubt by Nemyst · · Score: 4, Insightful

    Yes, because the word of the driver who totaled his car is also entirely reliable. I mean, it's not like he'd be liable if he crashed it himself, but could get a lot of money if Autopilot was the problem.

  3. Maybe the driver believed it was enabled? by davidwr · · Score: 2

    If the driver believed the autopilot was on when it was off, then we have to ask "why did he think it was on when it was off?"

    Was the driver not paying attention to the system, and just assumed it was on, or did the system lie and tell the driver that it was on when it wasn't?

    --
    Knowledge is how to play a game, intelligence is how to win, wisdom is knowing what game to play.
    1. Re:Maybe the driver believed it was enabled? by Archfeld · · Score: 3, Insightful

      Or maybe the driver just lied to cover up the fact that he was a poor driver who lost control and hit a guard rail. After all a car hitting a guard rail has never happened before driver assist was implemented. Nor has any driver attempted to cover their a$$ by lying to the cops about what they were doing that led up to the crash. I dropped the roach and was trying to find it when I slammed into the car ahead of me...err no that's not what happened

      --
      errr....umm...*whooosh* *whoosh* Is this thing on ?
    2. Re:Maybe the driver believed it was enabled? by robbak · · Score: 5, Interesting
      Possibly the driver, seeing the bridge or rail coming up and being uncomfortable with the approach speed, tapped the brake. This would have disabled the autopilot.

      Now, although disabling automatic systems on manual input has been the standard for as long as automatic systems have been available, I am beginning to wonder if it really is the right decision here. People seem to be turning it off without realising that they have done it.

      --
      Prediction for end of Universe #42: Fencepost error in Quantum_bogosort.cpp
    3. Re:Maybe the driver believed it was enabled? by Rei · · Score: 5, Insightful

      Exactly. Autopilot has to be the best excuse for wrecking your car ever invented. "No no, I didn't do it, the car did it itself. Really!"

      --
      Hourglass says she knows a kid in Iowa who grows up to be president.
    4. Re:Maybe the driver believed it was enabled? by KavyBoy · · Score: 2

      It's hard to not realize you've disabled it. First, there's a distinct two-tone chime. Second, if regenerative breaking is enabled at maximum (which I think most people do), the car slows down noticably unless you press the accelerator. And there's just a "feel" with the torque on the wheel or something. It's just hard to miss unless it's your first day using it or you're just not paying attention at all.

    5. Re:Maybe the driver believed it was enabled? by WalksOnDirt · · Score: 4, Informative

      The Tesla logs were reported as saying:

      Prior to the collision, Autosteer was in use periodically throughout the approximately 50-minute trip.

      The most recent such use ended when, approximately 40 seconds prior to the collision, the vehicle did not detect the driverâ(TM)s hands on the wheel and began a rapidly escalating set of visual and audible alerts to ensure the driver took proper control.

      When the driver failed to respond to 15 seconds of visual warnings and audible tones, Autosteer began a graceful abort procedure in which the music is muted, the vehicle begins to slow and the driver is instructed both visually and audibly to place their hands on the wheel.

      Approximately 11 seconds prior to the collision, the driver responded and regained control by holding the steering wheel, applying leftward torque to turn it, and pressing the accelerator pedal to 42%. Over 10 seconds and approximately 300m later and while under manual steering control, the driver drifted out of the lane, collided with a barrier, overcorrected, crossed both lanes of the highway, struck a median barrier, and rolled the vehicle.

      Now, you can believe this or not, but it doesn't match up with your hypothesis.

      --
      a,e,i,o,u and sometimes w and y (at be if of up cwm by)
    6. Re:Maybe the driver believed it was enabled? by sjames · · Score: 3, Insightful

      It can. Note that in an airplane, the pilot is expected to remain alert and at the controls while autopilot is engaged.

    7. Re:Maybe the driver believed it was enabled? by sjames · · Score: 2

      That's a hard one. Notably, we have yet to fix the problem of driver drifting off to sleep even when there isn't an autopilot to back them up.

    8. Re:Maybe the driver believed it was enabled? by WoOS · · Score: 2

      This can be fully in line with the driver thinking the autopilot being on-line. Potentially the driver was trained by the system over time that he just has to shortly move the steering wheel in answer to a "hands on wheel!" requests by the system to be allowed to take his hands off again for one/some minutes. Only this time he did it too late so the system did not re-engage.

      This is the problem of allowing long stretches of hands-off with only short stretches of hands-on because one originally promised "completely hands-off" to the customers.

  4. Wasn't me! by Anonymous Coward · · Score: 3, Insightful

    You can almost see this autopilot thing becoming an excuse whenever the car crashes. Me? Nah! It was the car! I swear it was the car! I am an always responsible driver! How can you dare saying that I am responsible for the car... it was the car itself I tell you! This AI is just bad I swear!

    So now that it is clear that *I* the driver has no fault on the crash, could you please not raise my insurance?... or prosecute me for killing that pedestrian, or running over that biker?... it was clearly not the alcohol... but the car!

    I'm glad they kept logs. Even if the logs are not 100% reliable, it is better than just the word of an honest driver that just happens to have someone to blame it onto.

  5. The one true metric should be.... by GerryGilmore · · Score: 4, Insightful

    ....is the per-mile-driven accident rate greater or less with Autopilot (or equivalent) enabled? Basically, it's a "perfect is the enemy of the good" situation whereby some folks seem to want to limit autonomous driving until it is 100% perfect when we all know that humans are far, far less reliable.

    1. Re:The one true metric should be.... by whoever57 · · Score: 3, Interesting

      ....is the per-mile-driven accident rate greater or less with Autopilot (or equivalent) enabled?

      For fatal accidents, the per-mile rate is lower with Autopilot enabled.

      But perhaps that doesn't tell the true story. Autopilot cannot be used in many situations, what if those situations are more dangerous? In other words, if the Autopilot can only be enabled on roads that are generally safer, then pure per-mile statistics are misleading.

      --
      The real "Libtards" are the Libertarians!
    2. Re:The one true metric should be.... by DerekLyons · · Score: 2

      ....is the per-mile-driven accident rate greater or less with Autopilot (or equivalent) enabled?

      For fatal accidents, the per-mile rate is lower with Autopilot enabled.

      But perhaps that doesn't tell the true story.

      No - it doesn't tell the whole story. The per-mile rate for non Autopilot vehicles is based on multiple billions of miles driven, where a single 'extra' death tomorrow would change the fourth or fifth decimal place. The per-mile rate for Autopilot is based on a much smaller sample size, a single death tomorrow would double the per-mile rate.

  6. Re:That radar really worked well in florida eh elo by Xenx · · Score: 4, Insightful

    If the accident was preventable, the driver should of prevented it. They should be paying attention to the road and be in a position to respond. If it wasn't preventable by the driver, then the system is working at least as well as the driver in that situation. Either way, the system isn't responsible.

    And for what it's worth, that doesn't mean the system couldn't/shouldn't be improved. It just means they didn't die because of the system.

  7. Re:its not musk's call by GerryGilmore · · Score: 4, Insightful

    So "victim's claims" are more accurate than "actual vehicle logs", eh? Your paranoia/hate is duly noted....

  8. Re:That radar really worked well in florida eh elo by saloomy · · Score: 2

    No. No one died because of technology. People died because they are not using the technology correctly. Tesla let the drivers know that the tech was in beta, Telsa let the drivers know to pay attention and still be alert and able to take control if the system fails to do exactly what it failed to do.

    Tech fails all the time, and thats OK, because programmers find the holes and improve the technology. Like people, its not full-proof. It gets better when people use it more. How many windows patches would there be if MS was still keeping it in a lab and trying to perfect it by themselves, to say nothing of FOSS. Use is the evolutionary process in software. Tesla Autopilot is already at human levels when statistically compared, and perhaps better even.

    So some one died. Thats sad. But the ROI on human life will be redeemed exponentially when the technology gets good enough to reduces fatalities on roadways to less than the hundreds of thousands we have today. Tesla and the rest of them are taking the dangers out of transportation for millions of users. They will make mistakes along the way, but the benefit is worth the cost here. Everyone who agrees to the Telsa Autopilot TOS agrees too, and if the users would follow the instructions more carefully, then the cost wouldn't be as high.

  9. Re:Beyond a doubt by Rei · · Score: 5, Interesting

    The Tesla logging system is not under investigation for being unreliable.

    Whether a hardware system is on or not is entirely different from any data that may be generated by it. There's a number of events required to enable autopilot, and all are logged.

    --
    Hourglass says she knows a kid in Iowa who grows up to be president.
  10. Re: I wouldn't be bragging quite so much... by Anonymous Coward · · Score: 3, Funny

    Wait. I thought the problem was that it DIDN'T miss the truck..

  11. Re:Beyond a doubt by PopeRatzo · · Score: 2

    I always trust log files

    I'm curious. Does the onboard driver assist computer phone home with log data to Tesla Motors? I'm pretty sure that in the case of an accident investigation, the car was impounded by the authorities. It would be pretty extraordinary if the highway patrol or accident investigators gave the information up to Tesla, considering Tesla didn't own the car.

    --
    You are welcome on my lawn.
  12. Re:Privacy? by Rei · · Score: 2

    Not me.

    I look forward to the day when car crashes are looked on the same as jet crashes. And one of the first goals with a passenger jet crash is to get the data off the flight recorders to crash investigators. The FAA takes the concept that crashes are simply not acceptable, just not "something that happens", and whatever caused a given crash doesn't just get marked "WILLNOTFIX" and closed. The latter is basically the situation with car accidents today. I strongly support, at least on an aspirational level, the concept that all data for every accident of significance is gathered, used to figure out what went wrong, and everything reasonably possible done to prevent that scenario from happening again.

    --
    Hourglass says she knows a kid in Iowa who grows up to be president.
  13. Re:That radar really worked well in florida eh elo by ThosLives · · Score: 3, Interesting

    ..the tech was in beta...

    This kind of thing - at the very least the finger pointing surrounding it - is why until now nobody put "beta" heavy machinery in the hands of the general public.

    The general public should never be assumed to use things as designed - not all product liability lawsuits are as frivolous as they are sometimes portrayed.

    I would probably find Tesla negligent just on the grounds that they are assuming people won't abuse (or even simply misuse!) the feature. Waivers notwithstanding - it would be interesting to see those in court, because I guarantee just about everyone who signed one would have to say "I just signed it to get the shiny, I don't know what it said" if they were being honest. This means there is no evidence of expectation in the general public that these things are "beta".

    Put another way: you can call something "beta" all you want in theory, but if you're selling it to the general public, it ain't in practice.

    --
    "There are a dozen opinions on a matter until you know the truth. Then there is only one." - CS Lewis (paraprhase)
  14. A Safer Cruise Control by SmaryJerry · · Score: 4, Insightful

    I love that the feature wasn't even on. Go ask any insurance adjuster and let them tell you if people lie about accidents. But even if it was, this feature is just a cruise control that also keeps you in your lane and might brake when an object is in your way. It is literally a far safer cruise control than any other vehicle. This doesn't mean you can sleep while using it, same as other cruise control. If I told you I had a helmet that made injuries to the brain 50% less likely, that wouldn't mean you can use it to dive off a building head first. Using products in ways other than intended is not the fault of Tesla.

  15. Better than I thought though by s.petry · · Score: 2

    My Drivers Ed teacher thir*cough* years ago had a story, I don't know if it was true but it was funny. A kid he was teaching was told to put on the cruise control during the freeway portion. He did, and took his hands off the wheel. He thought Cruise Control was an auto-pilot.

    He ended up passing, but they stopped having people use cruise control.

    --

    -The wise argue that there are few absolutes, the fool argues that there are no probabilities.

  16. Re:Beyond a doubt by lucm · · Score: 3, Insightful

    So what. It's not an either/or situation. Both sides have an incentive to spin things, that's not by itself an argument for the other side.

    Yes, there are dishonest people out there. But also anyone with experience in mass production will tell you that real life has a tendency to make flukes and quirks happen sooner than later. There's just too many factors in the production line and in the end user environment to account for all possible defects. Cars are not simple apps running on heroku, there's a physical and mechanical element to consider.

    Ever heard of problems with early MAP sensors back when they were installed inside the cab? Difference of temperature in winter tended to cause condensation in the line, but by the time the car was on the lift at the dealership, the condensation was gone. In hindsight the problem is obvious, but back then the "conclusive tests" made by the manufacturer made them treat honest people like crooks or idiots.

    --
    lucm, indeed.
  17. Re:Beyond a doubt by thegarbz · · Score: 2

    So what. It's not an either/or situation. Both sides have an incentive to spin things, that's not by itself an argument for the other side.

    Yep, that's when you look at history. On the one side we have a driver who we have no history on talking to the police.
    On the other side you have a company who voluntarily reported that an autopilot feature was activated during a crash to regulators, who while they stand to lose a lot have shown a history of honesty.

    Who do you believe?

  18. Re:its not musk's call by thegarbz · · Score: 2

    Thanks for your generalisation about a company who's CEO voluntarily disclosed to regulators who weren't investigating the autopilot feature that the autopilot feature has been the result of 2 crashes prior. Clearly they are not to be trusted.

    Your comment shows us a great deal about your ability to use your brain.

  19. Re:Beyond a doubt by Lumpy · · Score: 2

    Yes because WSJ is the epitome of engineering information....

    Got any quotes from Fox news as well?

    --
    Do not look at laser with remaining good eye.
  20. "First of three.. no fatalities" = FUD by Theaetetus · · Score: 5, Interesting

    In response to the third reported Autopilot crash, which was the first of three where there were no fatalities

    The first crash in Florida was the guy who got killed going under the truck while watching his DVD.
    The second crash was a gallery owner in Detroit and he and his passenger survived without any injuries .
    The third crash - the one apparently without autopilot - hit a guard rail in Montana. "The two occupants walked away without major injuries."

    I don't know why this "fatalities in two crashes" myth is so pernicious. It was also falsely claimed in this Slashdot story on the third crash last Monday. But all of the linked articles are absolutely clear that there's been only one fatality, so it's not like the various submitters are just getting bad information from the media. Instead, the Subbys appear to be making up the second fatality out of nothing.

    A more skeptical person than me would wonder if someone shorted TSLA.

  21. Re:Beyond a doubt by geoskd · · Score: 2

    Lack of data isn't a good argument for not trusting someone.

    Yes it is. There is an age old adage the goes "Trust isn't given, its earned". There is 10,000 years of wisdom in that expression.

    --
    I wish I had a good sig, but all the good ones are copyrighted
  22. Re:That radar really worked well in florida eh elo by geoskd · · Score: 4, Insightful

    "well it's only in beta". In which case it should never have been out on the public road.

    Engineering doesn't work that way. Driving is a complicated enough task that working out the majority of the kinks in an autopilot system will requires billions of miles of road testing to start approaching "perfect". That would require hundreds of billions of dollars in testing costs. In the mean time, tens of thousands of drivers per year get killed in avoidable accidents. At some point, you have to stop and realize that the cost of "perfect" far exceeds the cost of getting it on the road in an unpolished state. You also have to realize that some part of that cost will be in human lives (on both sides of the equation).

    If the airline manufacturers had to get it perfect, we would still be testing aircraft today, and the only air travel we would have would be blimps. Instead, we have the NTSB and FAA, which work to ensure that air traffic safety improves as time goes by. There is no guarantee that you won't die tomorrow in a horrible plane crash. The guarantee you do get is that the engineers will figure out why it happened and take steps to prevent further occurrences. The same approach has been true for automobiles, and cars today are the safest they have ever been, but like controlled flight, the most dangerous component is the pilot/driver. Autopilots on planes have been improving rapidly over the last 20 years, and even the pilot induced crashes are less frequent than they used to be. The sooner we can begin this iterative improvement process for over the road vehicles, the sooner we can start to see a reduction in auto accidents. Even if autonomous vehicles are worse today, if they end up reducing automobile fatalities by as little as 10% per year when they are mature, then having them kill an excess of 15,000 people over that 5 year span while they are still being "perfected", is a break even as far as human lives are concerned. In that time, Tesla has killed an average of 4 per year, or 20 lives paid for the technology that will save thousands per year when it is fully mature. You would have to be an idiot not to take that deal.

    Given the stakes, Congress should take steps today to structure the process in exactly the same way as the very successful air safety system the whole world uses to reduce airline crashes. As part of that, I would take it one step further, and automatically indemnify any auto manufacturer that releases their system designs for public scrutiny.

    --
    I wish I had a good sig, but all the good ones are copyrighted