Shop OBEX P1 Docs P2 Docs Learn Events
First Self-Driving Car Fatality — Parallax Forums

First Self-Driving Car Fatality

GordonMcCombGordonMcComb Posts: 3,366
edited 2018-03-19 17:57 in Robotics
It took Uber to screw it up for everyone. One of their test self-driving cars, which had a human control riding in it but not driving at the time, killed a pedestrian earlier today.

http://money.cnn.com/2018/03/19/technology/uber-autonomous-car-fatal-crash/index.html

I have always said it takes just one bad incident to stop or greatly curtail new technology such as self-driving vehicles. The public's perception of technology is always tenuous. The "Buckminster Fuller effect" is when an emerging technology suffers a public relations fiasco. and irrepairably harms the public trust. In Fuller's case it was the Dymaxion car, which was involved in a fatal accident largely unrelated to its technology. The design of the car, and many elements of it, were just abandoned because of the bad publicity.

We all knew self-driving wasn't perfect, but it seems to me hubris played a large role in this. What's the point of having a human control during the tests if the human isn't going to intercede? IMO, Uber needs to go back to square one, take all of their cars off the road for foreseeable future, and let other developers of autonomous vehicles proceed with the caution that has so far kept this sort of accident from occurring.
«1

Comments

  • I continue to wonder how self-driving cars will handle unusual conditions. Closed lanes due to rode construction. Flag men that handle traffic flow. Immigration check points, such as the ones we have in south Texas. A 2x4 laying in the road that fell off a truck a few miles ahead. An odd sound in the engine compartment that might indicate an engine failure may happen soon. A dump truck with loose gravel trickling out of it. An aggressive driver cutting in front of it while another driver is tailgating it. It will be interesting to see how the designers will handle these situations that human drivers have to contend with all the time.
  • ercoerco Posts: 20,244
    Uber's only other accident was also in Tempe AZ. What's up?

    Uber has previously grounded its vehicles while investigating a crash. In 2017, Uber briefly pulled its vehicles from roads after an Uber self-driving vehicle in Tempe landed on its side.
  • Not enough details about the accident have been reported, so it's hard to know if this was caused by a sensor or AI failure, or if the pedestrian suddenly walked out in traffic (intentionally or otherwise). Isaac Newton is still the driver in all of these; if someone dashes out in front of a car, an autonomous vehicle will have a faster reaction time, but inertia is still inertia.

    While I believe in the future of self-driven cars, MUCH more testing needs to be done, and their first rollout must be in controlled environments -- self-driven trams, busses on special bus lanes, inter-campus transportation, and so on. Even golf carts can be an effective test bed for autonomous vehicles, and far safer for the public.
  • The_MasterThe_Master Posts: 199
    edited 2018-03-20 01:17
    I'm calling it right now: There will be criminal charges filed for this accident.
  • My condolences to the victim and her family.
    While I believe in the future of self-driven cars, MUCH more testing needs to be done, and their first rollout must be in controlled environments -- self-driven trams, busses on special bus lanes, inter-campus transportation, and so on. Even golf carts can be an effective test bed for autonomous vehicles, and far safer for the public.

    I'll second this. It's the classic "walk before run" risk mitigation.
  • I'm not sure there will be any criminal liability against Uber, but certainly Uber will be sued by the woman's family, and likely settle out of court. I don't know about the test driver. That'll depend on what he was doing at the time of the accident.

    What they're saying now is that the woman was crossing the road walking her bicycle some 60 yards from an intersection -- IOW jaywalking. Human drivers know to watch for this sort of thing and might have slowed down. But (armchair engineering here), did the Uber vehicle have an effective AI for that scenario? Police are reporting evidence collected so far don't show "significant signs" that it slowed down.

    Situations like someone suddenly crossing into traffic happen every day. Often they are hit, but aware drivers may swerve and brake to avoid a collision. You have to wonder what the Uber car was programmed for. I guess we'll find that out over time.
  • There's an update saying the onboard video shows the pedestrian would have been hard to avoid.

    That seems typical in a lot of pedestrian incidents. Maybe we need self walking legs next.

  • Yes, I read that report. It mentioned she "came out of the shadows," which is a curious statement as shadow and darkness is a limitation of a human driver, not an autonomous driver that uses LIDAR and other non-visible light sensors.

    Here's the thing: there's an average of slightly more than 1 fatality per 100 million driven miles. So far driverless cars have racked up maybe 10 million actual road miles, which is just a tenth of what's needed to even begin assessing the chances of fatal accidents.

    All it takes is ONE fatality in the early years to set everything back, years maybe even decades. This is why I think it's wholly inappropriate to rush the technology onto open roads -- it's not what the developers think (remember Google still thinks it's okay to put a camera on your eyeglasses to record everything you see), it's the trust the public puts into it. It's all about the public buying into the concept, not the developers selling it.


  • Wait a second here. So the pedestrian was jay walking? If so, then Uber's liability just went out the window. There are cross walks for a reason. Not that I don't doubt for a second that lawyers, media, etc, will try to spin it all for the woman who was hit, and against the technology, but...
  • Phil Pilgrim (PhiPi)Phil Pilgrim (PhiPi) Posts: 23,514
    edited 2018-03-21 02:19
    At this juncture of AI, designing self-driving cars and thinking they're road-ready is a total exercise in hubris. Although, on average, self-driven cars may be safer than those driven by humans -- particularly those humans prone to distraction -- humans are still better at fielding the edge cases and probably will be for years to come.

    -Phil
  • Cluso99Cluso99 Posts: 18,066
    edited 2018-03-21 09:41
    Humans are also capable, and do, deliver the edge cases in the first place - requiring all those edge case reactions!!!
  • DrPop wrote: »
    Wait a second here. So the pedestrian was jay walking? If so, then Uber's liability just went out the window.

    Not necessarily. Liability isn't a 0% or 100% thing. She may be at more than 51% fault, but that doesn't mean Uber won't be sued in civil court. Because of the negative publicity the most likely outcome is a confidential settlement. No trial attorney will let Uber slide on the fact that it only has about 2 million road miles under its belt. There's a strong argument that not enough testing was performed before the car was placed in public use. This is not a scenario Uber or anyone in the industry wants.

  • Cluso99 wrote:
    Humans are also capable, and do, deliver the edge cases in the first place - requiring all those edge case reactions!!!
    Perhaps pedestrians should be licensed, too, after taking a rigorous exam. Then they could be fined for walking without a license and even distracted walking! :)

    -Phil
  • kwinnkwinn Posts: 8,697
    Cluso99 wrote:
    Humans are also capable, and do, deliver the edge cases in the first place - requiring all those edge case reactions!!!
    Perhaps pedestrians should be licensed, too, after taking a rigorous exam. Then they could be fined for walking without a license and even distracted walking! :)

    -Phil

    LOL. After years of downtown driving dealing with headphone zombies, texting twits, and crazed cyclist I am warming to the idea. Perhaps a more appropriate response would be to enforce existing laws and enhance them as needed. A boost to employment and government income along with increased safety. What more could we ask for?
  • kwinn wrote: »
    Cluso99 wrote:
    Humans are also capable, and do, deliver the edge cases in the first place - requiring all those edge case reactions!!!
    Perhaps pedestrians should be licensed, too, after taking a rigorous exam. Then they could be fined for walking without a license and even distracted walking! :)

    -Phil

    LOL. After years of downtown driving dealing with headphone zombies, texting twits, and crazed cyclist I am warming to the idea. Perhaps a more appropriate response would be to enforce existing laws and enhance them as needed. A boost to employment and government income along with increased safety. What more could we ask for?

    Or, double down on autonomous vehicles. If there are more autonomous vehicles on the road, things will resolve themselves one way or the other! :-D
  • Searth wrote:
    Or, double down on autonomous vehicles. If there are more autonomous vehicles on the road, things will resolve themselves one way or the other!
    ... until the pedestrian masses organize with their tacks and nail strips, declaring all-out war against their self-driving robot overlords.

    -Phil
  • ercoerco Posts: 20,244
    It would be very interesting if Autocars forced people to walk & bike more carefully/lawfully, knowing that they give up their legal rights if they jaywalk or blow through a red light.

    Then again, there are the Three Laws of Robotics to contend with.
  • erco wrote: »
    Then again, there are the Three Laws of Robotics to contend with.

    Ah, but that only works if the robot knows its action (or inaction) can affect the outcome. As mentioned earlier, even robots can't circumvent the laws of physics. So, now, if a robot must take into account the possibility of violating one of the laws in the future for an action taken in the past, then the robot will never take any action at all. Unless "taking no action" is itself considered an action, in which case, the robot will go insane. Not too surprisingly, Asimov delved into this scenario as well (the short story "Liar!" comes to mind).

  • erco wrote: »
    Then again, there are the Three Laws of Robotics to contend with.
    Very problematic or as Deep Thought would say:

    1.Hitchhiker 25: (41) Deep Thought paused for a moment's reflection. (42) "Tricky," he said finally."

  • Cluso99Cluso99 Posts: 18,066
    Cluso99 wrote:
    Humans are also capable, and do, deliver the edge cases in the first place - requiring all those edge case reactions!!!
    Perhaps pedestrians should be licensed, too, after taking a rigorous exam. Then they could be fined for walking without a license and even distracted walking! :)

    -Phil

    How many pedestrians just walk out without looking? I have seen tons, mostly the younger ones. They make sure they don't look at you when you slam on the brakes. Many have earplugs and phones out. And who do they blame if they get hit?

    BTW A number of cars have had auto breaking for a while now.

    A number of manufacturers have announced all their current models are now fully upgradable to self-driving. Lane departure warning, blind spot warning, auto braking, etc are often now standard in reasonable models now. I happen to know Volvo are very well advanced, and VW have had lots of extras for some time too. Mazda MX9 has all these too.
  • ercoerco Posts: 20,244
    Cluso99 wrote: »
    How many pedestrians just walk out without looking? I have seen tons, mostly the younger ones. They make sure they don't look at you when you slam on the brakes. Many have earplugs and phones out. And who do they blame if they get hit?

    I think some of these "entitled" oblivious kids are hoping to get hit by a car and get a cash settlement.

    Sure beats working.

  • In years to come people will say with astonishment-
    "you mean to say people and cars moved within feet of each other and there were no barriers!!!- unbelieveable!- sure were primitive back then".

    Dave
  • If the situation is unavoidable, at least the autonomous vehicles can provide the judge with concrete evidence.

    None of these vehicles should be classified as self-driving. There should be a name for what we have now.

    SSD (Semi Self Driving)?
  • After looking at the dashcam video, it's clear that the car made no attempt to stop or avoid the pedestrian. It appears that the pedestrian wasn't visible until about 1.5 seconds before being hit. This wouldn't have been enough time for a human to stop the car, but it may have been possible to swerve to the left to avoid hitting the pedestrian. If this would have been during the daylight a human would have seen the person with enough time to avoid hitting her. It seems like the car's LIDAR should have detected the pedestrian with enough time to avoid the accident. Something must have failed. The software needs to be able to detect if there is a sensor failure, and it should pull the car over and stop it if that's the case.
  • xanaduxanadu Posts: 3,347
    edited 2018-03-22 00:53
    The dashcam is not low light sensitive. The safety driver should have seen her. Safety driver needs to find a new line of work.
  • (word of warning: the dashcam video is unsettling.)
  • xanadu wrote: »
    The dashcam is not low light sensitive. The safety driver should have seen her. Safety driver needs to find a new line of work.
    Yes, the safety driver should have been paying attention, and maybe the accident could have been avoided. However, it's not clear that the driver could have reacted quickly enough to prevent the accident.

  • At the very least, it seems like they could have delayed night driving (where the safety driver is guaranteed to have limited visibility) until it was fully proven in daylight.
  • Video is totally different as described above. Everyone should go watch it. It is not the least bit disturbing, moment of impact not shown. (at least on the ABCnews version)

    A person does not 'jump out' in front of the car, they are walking across multiple lanes of traffic.

    Dashcam video: It is nighttime, car is probably doing 45mph on a very, very well lit road. Moron is walking a bike across multiple lanes of traffic. Despite it being a major road with well-beamed cars roaring at them, he is looking in opposite direction.

    Because it is a well lit street and not a dark 70 mph freeway, any one of us would have easily seen the person and slowed down/stopped. Perhaps this is why the pedestrian felt entitled to be acting like this in the first place.

    Maybe the video SORTA looks 'dark' because of the low res video you saw.
  • 'Just watched the dashcam video. The bicyclist/pedestrian was obviously oblivious to oncoming traffic. However, it seems that the LIDAR should've detected her and slammed on the brakes. But I'm sure there's more to it than what the video shows, so I won't even try to second-guess the situation. After a data dump and an NTSB investigation, there's sure to be more enlightening news going forward.

    -Phil
Sign In or Register to comment.