3 ms·

That’s only because there’s no light on her until the end. The car hardware should have seen her. If it doesn’t work in the dark it shouldn’t be on the road im

That’s only because there’s no light on her until the end. The car hardware should have seen her. If it doesn’t work in the dark it shouldn’t be on the road imho.

lolsal 9y ago
> That’s only because there’s no light on her until the end. Yes, quiet true. > The car hardware should have seen her. If it doesn’t work in the dark it shouldn’t be on the road imho. I tend to agree. I think unfortunately this video indicates that the car did not perform worse than a human. It's still valuable to have cars that perform better than humans, even if they aren't perfect.
- foobarbazetc 9y ago
  What do you think based on this video: https://twitter.com/brianawhitney/status/976591851384745984?s=21 https://twitter.com/brianawhitney/status/976591851384745984?... The woman with the bike wasn’t moving fast at all. I dunno...
  - lhorie 9y ago
    Isn't that the exact same video in poorer quality? I think the question now is how well a LIDAR is supposed to work in the dark. I keep hearing people saying that the hardware should've "seen" her, but speaking purely in terms of physics, how exactly is that supposed to work? Doesn't night vision still require a minimum threshold of light in order to work? How much light?
    - brokenmachine 9y ago
      LIDAR sends out pulsed laser beams to measure the distance to things. It should work in complete darkness.
      - lhorie 9y ago
        > It should work in complete darkness If that's true, this would certainly suggest a big failure from the LIDAR system in this incident. I wonder how that claim is tested. Would city lights interfere with testing (similar to how you can't see stars in the middle of cities)?
        brokenmachine 9y ago
        Someone in another comment rightly mentioned that LIDAR might just be one input with visual/radar being others. So it may have ignored the LIDAR because of whatever was coming from the other inputs. It will be interesting to find out what actually happened. It is clearly a major malfunction or deficiency in the sensors, no doubt. For a self-driving car to not even apply the brakes at all in this situation is inexcusable.
  - nemothekid 9y ago
    Even in that over-exposed video, at 35mph, I think you would have to be extremely lucky to not hit her. From her coming into view to collision is ~700ms. Subtract average human reaction time (250ms), and that leaves a little less than half a second to swerve. A human would have made this mistake, but I don't know enough about LIDAR to understand if a computer should have made this mistake. Is night vision poor on these machines?
    - JustSomeNobody 9y ago
      Then slow down. If you can't stop or swerve when something becomes visible to you then you're driving too fast regardless of the posted speed limit. Period.
- brokenmachine 9y ago
  > I tend to agree. I think unfortunately this video indicates that the car did not perform worse than a human. It's still valuable to have cars that perform better than humans, even if they aren't perfect. I disagree. I think this video shows that the car performs much worse than I would expect a human to. I have my own dashcam videos at night that look very similar to this, and I definitely think I would not have hit a large object such as a pedestrian walking a bicycle on a straight road such as this in a 35mph zone.
thrill 9y ago
Well, if we're demanding robocars that are better than people at this task, then that is a higher standard than we currently enjoy.
- JustSomeNobody 9y ago
  Literally everyone is constantly saying self driving cars will be better than humans. So, yeah, that's the standard.
Figs 9y ago
> That’s only because there’s no light on her until the end. It seems like the car is out driving its headlights. Assuming a road line is about 10 feet (as is typical), only about 20~30 feet of distance is illuminated... but at 30mph you need ~45ft for an average car to break, and at 40mph you need ~80ft. (According to [1] -- ignoring reaction time.) Is there something wrong with the headlights? Is the video just that bad? (And if so, is it even used for decision making by the car?) [1] http://www.government-fleet.com/content/driver-care-know-your-stopping-distance.aspx http://www.government-fleet.com/content/driver-care-know-you...

That’s only because there’s no light on her until the end. The car hardware should have seen her. If it doesn’t work in the dark it shouldn’t be on the road imho.

lolsal 9y ago

> That’s only because there’s no light on her until the end. Yes, quiet true. > The car hardware should have seen her. If it doesn’t work in the dark it shouldn’t be on the road imho. I tend to agree. I think unfortunately this video indicates that the car did not perform worse than a human. It's still valuable to have cars that perform better than humans, even if they aren't perfect.

foobarbazetc 9y ago
What do you think based on this video: https://twitter.com/brianawhitney/status/976591851384745984?s=21 https://twitter.com/brianawhitney/status/976591851384745984?... The woman with the bike wasn’t moving fast at all. I dunno...
- lhorie 9y ago
  Isn't that the exact same video in poorer quality? I think the question now is how well a LIDAR is supposed to work in the dark. I keep hearing people saying that the hardware should've "seen" her, but speaking purely in terms of physics, how exactly is that supposed to work? Doesn't night vision still require a minimum threshold of light in order to work? How much light?
  - brokenmachine 9y ago
    LIDAR sends out pulsed laser beams to measure the distance to things. It should work in complete darkness.
    - lhorie 9y ago
      > It should work in complete darkness If that's true, this would certainly suggest a big failure from the LIDAR system in this incident. I wonder how that claim is tested. Would city lights interfere with testing (similar to how you can't see stars in the middle of cities)?
      - brokenmachine 9y ago
        Someone in another comment rightly mentioned that LIDAR might just be one input with visual/radar being others. So it may have ignored the LIDAR because of whatever was coming from the other inputs. It will be interesting to find out what actually happened. It is clearly a major malfunction or deficiency in the sensors, no doubt. For a self-driving car to not even apply the brakes at all in this situation is inexcusable.
- nemothekid 9y ago
  Even in that over-exposed video, at 35mph, I think you would have to be extremely lucky to not hit her. From her coming into view to collision is ~700ms. Subtract average human reaction time (250ms), and that leaves a little less than half a second to swerve. A human would have made this mistake, but I don't know enough about LIDAR to understand if a computer should have made this mistake. Is night vision poor on these machines?
  - JustSomeNobody 9y ago
    Then slow down. If you can't stop or swerve when something becomes visible to you then you're driving too fast regardless of the posted speed limit. Period.
brokenmachine 9y ago
> I tend to agree. I think unfortunately this video indicates that the car did not perform worse than a human. It's still valuable to have cars that perform better than humans, even if they aren't perfect. I disagree. I think this video shows that the car performs much worse than I would expect a human to. I have my own dashcam videos at night that look very similar to this, and I definitely think I would not have hit a large object such as a pedestrian walking a bicycle on a straight road such as this in a 35mph zone.

thrill 9y ago

Well, if we're demanding robocars that are better than people at this task, then that is a higher standard than we currently enjoy.

JustSomeNobody 9y ago
Literally everyone is constantly saying self driving cars will be better than humans. So, yeah, that's the standard.

Figs 9y ago

> That’s only because there’s no light on her until the end. It seems like the car is out driving its headlights. Assuming a road line is about 10 feet (as is typical), only about 20~30 feet of distance is illuminated... but at 30mph you need ~45ft for an average car to break, and at 40mph you need ~80ft. (According to [1] -- ignoring reaction time.) Is there something wrong with the headlights? Is the video just that bad? (And if so, is it even used for decision making by the car?) [1] http://www.government-fleet.com/content/driver-care-know-your-stopping-distance.aspx http://www.government-fleet.com/content/driver-care-know-you...