4 ms·

In all seriousness though, what would happen in this case? Do self-driving cars have emergency stop procedures? And how safe are they? This is still a serious c

by vlunkr 12y ago

In all seriousness though, what would happen in this case? Do self-driving cars have emergency stop procedures? And how safe are they? This is still a serious concern about self-driving cars for me. Especially since digital image recognition still isn't that great. What if the car sees a tumbleweed that it mistakes for a boulder and puts it into emergency stop mode?

munificent 12y ago
> Do self-driving cars have emergency stop procedures? Yes, of course! Just staying on a road is a piece of cake. Most of the research for self-driving cars is dealing with unexpected cars and pedestrians. > And how safe are they? According to the article, 1.7m miles driven without at at-fault accident. Self-driving cars drive very conservatively. > What if the car sees a tumbleweed that it mistakes for a boulder and puts it into emergency stop mode? Probably the same thing that happens when humans make visual mistakes: it would slow down and stop. Of course, the self-driving car is looking at the tumbleweed in the visible spectrum, using LIDAR, maybe IR and probably some other sensors. We just have our squishy eyeballs.
- vlunkr 11y ago
  I guess what I'm asking is are the emergency stop procedures safe? If you're going 75mph, how much distance does it need to stop without rolling or harming the passengers?
phaemon 12y ago
> Especially since digital image recognition still isn't that great. I believe digital image recognition is better than human, and a lot faster to boot (hah!).
- vlunkr 11y ago
  You should take a look at this: http://fastml.com/what-you-wanted-to-know-about-ai/ http://fastml.com/what-you-wanted-to-know-about-ai/. Digital image recognition isn't that good
  - phaemon 11y ago
    I read over it, but there are a couple of problems with that. Firstly, I never said that computers understand the images, just that the recognition is better than human. When they get it wrong, they do tend to get it wrong in ways that look ludicrous to people. Secondly, Stanford didn't win the 2014 ImageNet challenge: GoogLeNet did. However, I was still wrong. Have a look at the summary paper from ImageNet: http://arxiv.org/abs/1409.0575 http://arxiv.org/abs/1409.0575 Page 31 states: "Annotator A1 evaluated a total of 1500 test set images. The GoogLeNet classification error on this sample was estimated to be 6.8% (recall that the error on full test set of 100,000 images is 6.7%, as shown in Table 7). The human error was estimated to be 5.1%." There are some bits there where GoogLeNet did better than human, but it was usually on specific classifications (what breed of dog is that, for example). Generally, a human is slightly better. It's worth noting, however, that the error rate was halved from 2013's winners. I reckon the 2015 error rate will beat human, and in following years there will just be no contest. Just like chess programs have gone far beyond human capabilities. And, of course, with cars we're talking about having multiple images plus depth perception, which makes it a lot easier for the machine.

by vlunkr 12y ago

munificent 12y ago

> Do self-driving cars have emergency stop procedures? Yes, of course! Just staying on a road is a piece of cake. Most of the research for self-driving cars is dealing with unexpected cars and pedestrians. > And how safe are they? According to the article, 1.7m miles driven without at at-fault accident. Self-driving cars drive very conservatively. > What if the car sees a tumbleweed that it mistakes for a boulder and puts it into emergency stop mode? Probably the same thing that happens when humans make visual mistakes: it would slow down and stop. Of course, the self-driving car is looking at the tumbleweed in the visible spectrum, using LIDAR, maybe IR and probably some other sensors. We just have our squishy eyeballs.

vlunkr 11y ago
I guess what I'm asking is are the emergency stop procedures safe? If you're going 75mph, how much distance does it need to stop without rolling or harming the passengers?

phaemon 12y ago

> Especially since digital image recognition still isn't that great. I believe digital image recognition is better than human, and a lot faster to boot (hah!).

vlunkr 11y ago
You should take a look at this: http://fastml.com/what-you-wanted-to-know-about-ai/ http://fastml.com/what-you-wanted-to-know-about-ai/. Digital image recognition isn't that good
- phaemon 11y ago
  I read over it, but there are a couple of problems with that. Firstly, I never said that computers understand the images, just that the recognition is better than human. When they get it wrong, they do tend to get it wrong in ways that look ludicrous to people. Secondly, Stanford didn't win the 2014 ImageNet challenge: GoogLeNet did. However, I was still wrong. Have a look at the summary paper from ImageNet: http://arxiv.org/abs/1409.0575 http://arxiv.org/abs/1409.0575 Page 31 states: "Annotator A1 evaluated a total of 1500 test set images. The GoogLeNet classification error on this sample was estimated to be 6.8% (recall that the error on full test set of 100,000 images is 6.7%, as shown in Table 7). The human error was estimated to be 5.1%." There are some bits there where GoogLeNet did better than human, but it was usually on specific classifications (what breed of dog is that, for example). Generally, a human is slightly better. It's worth noting, however, that the error rate was halved from 2013's winners. I reckon the 2015 error rate will beat human, and in following years there will just be no contest. Just like chess programs have gone far beyond human capabilities. And, of course, with cars we're talking about having multiple images plus depth perception, which makes it a lot easier for the machine.