4 ms·

Elsewhere on HN, someone posted this video: https://www.youtube.com/watch?v=UZ1XLqc5IUg https://www.youtube.com/watch?v=UZ1XLqc5IUg This was my comment reply:

Elsewhere on HN, someone posted this video: https://www.youtube.com/watch?v=UZ1XLqc5IUg https://www.youtube.com/watch?v=UZ1XLqc5IUg This was my comment reply: Wow. That guy is going nineteen miles per hour on a simple road in daylight and the car literally crossed to the wrong side of the road in nine seconds! That's an absolute disaster.

gambiting 10y ago
As much as I would love to have a self-driving car(and I would!) I honestly shake my head at anyone saying that we will have fully autonomous vehicles in the next 5-10 years. Unless the roads are fully standardised and beacons placed every few meters, this is literally the image recognition problem all over again. In the 60s scientists thought that with the power of computers they will "solve" image recognition completely within a few months. It's 2017 and our best of the best algorithms and neural networks are stupidly bad at things they haven't been specifically trained to recognise, and even then they make catastrophic mistakes. And then we want to put essentially this tech on a car, where our roads are bad, signs are faded or straight up wrong, where you can have any sort of weather conditions obscuring the view and/or lidar, plus the whole thing has to be so reliable that you can put it on a consumer product......within 10 years? If anything, we will get a better version of Tesla Autopilot within 10 years, but full autonomy? Try 50 years.
- Nomentatus 10y ago
  Humans aren't all that great at driving safely and consistently either. Autonomous vehicles just have to be better than the average (not terrific) driver, not perfect; and progress recently beggars the progress in previous generations. I'm placing my money on sooner rather than later in this case.
  - owenversteeg 10y ago
    That would make logical sense, yes. But humans aren't logical, and regulators won't approve autonomous vehicles until they make accidents exceedingly rare. Also, even the worst driver I know drives better than the Tesla in that video. Seriously, watch the video if you haven't already. I think Google/Waymo are really going to be the winners here.
  - gambiting 10y ago
    You see, this is where I disagree. "better than a human" was never a valid defense against a machine failing and killing someone. Why would it be different with autonomous cars? Let's say we have a fully 100% autonomous car, where you don't have to pay attention to the road. It crashes, injures or kills someone - who is responsible? If a human did that, then we call it "an accident" and your insurance company foots the bill. With an autonomous car, sure, an insurance company will pay for the damages - but they will demand that a problem is fixed or they won't insure those cars anymore. So anything less than 100% perfect is a huge liability for manufacturers, and we are very very far off 100% perfect. I cannot imagine a world where autonomous cars have minor/major accidents but we just brush it off as "meh, whatever - humans have a higher rate of accidents so it's fine". Or maybe think about it this way. Introduction of autopilot in airplanes resulted in a major improvement in safety, the number of plane crashes has literally plummeted(sorry) after its introduction. But it hasn't eliminated crashes entirely - and every time there is a crash, Boeing/Airbus will ground their entire fleets to investigate what went wrong, if there is a perceived fault then thousands of planes have to be updated before they are safe to fly. Now bring that analogy over to cars - small, personal transportation vehicles, where profit margins really aren't that fantastic. Which manufacturer will risk releasing a product where they have to keep updating and maintaining the autopilot for the lifetime of the car, and for which they are liable every time it fails? Not to mention that planes are held to ridiculous safety standards, while cars are not(so far, I haven't heard an answer to what happens if a bird shits on the lidar and you don't clean it - it's an instrument with almost chirurgical precision which requires similar level of maintenance, and I know people who buy a car and then don't even wash it once in 3 years of owning it). I'm not saying we won't get there. I'm saying that I don't see any chance that this technology will be ready in 5-10 years, and if anyone is foolish enough to sell an "autonomous" car in the next 10 years, they will set back our progress massively because the car will fail, and there will be public and legislatory backlash.
    - Nomentatus 10y ago
      We get used to things that seem unsafe but are safe very quickly according to history. Elevators, for example. When I was young they generally had human operators, but now they're all autonomous, have been for a while, fail sometimes here and there but nobody seems hugely exercised about it.

gambiting 10y ago

As much as I would love to have a self-driving car(and I would!) I honestly shake my head at anyone saying that we will have fully autonomous vehicles in the next 5-10 years. Unless the roads are fully standardised and beacons placed every few meters, this is literally the image recognition problem all over again. In the 60s scientists thought that with the power of computers they will "solve" image recognition completely within a few months. It's 2017 and our best of the best algorithms and neural networks are stupidly bad at things they haven't been specifically trained to recognise, and even then they make catastrophic mistakes. And then we want to put essentially this tech on a car, where our roads are bad, signs are faded or straight up wrong, where you can have any sort of weather conditions obscuring the view and/or lidar, plus the whole thing has to be so reliable that you can put it on a consumer product......within 10 years? If anything, we will get a better version of Tesla Autopilot within 10 years, but full autonomy? Try 50 years.

Nomentatus 10y ago
Humans aren't all that great at driving safely and consistently either. Autonomous vehicles just have to be better than the average (not terrific) driver, not perfect; and progress recently beggars the progress in previous generations. I'm placing my money on sooner rather than later in this case.
- owenversteeg 10y ago
  That would make logical sense, yes. But humans aren't logical, and regulators won't approve autonomous vehicles until they make accidents exceedingly rare. Also, even the worst driver I know drives better than the Tesla in that video. Seriously, watch the video if you haven't already. I think Google/Waymo are really going to be the winners here.
- gambiting 10y ago
  You see, this is where I disagree. "better than a human" was never a valid defense against a machine failing and killing someone. Why would it be different with autonomous cars? Let's say we have a fully 100% autonomous car, where you don't have to pay attention to the road. It crashes, injures or kills someone - who is responsible? If a human did that, then we call it "an accident" and your insurance company foots the bill. With an autonomous car, sure, an insurance company will pay for the damages - but they will demand that a problem is fixed or they won't insure those cars anymore. So anything less than 100% perfect is a huge liability for manufacturers, and we are very very far off 100% perfect. I cannot imagine a world where autonomous cars have minor/major accidents but we just brush it off as "meh, whatever - humans have a higher rate of accidents so it's fine". Or maybe think about it this way. Introduction of autopilot in airplanes resulted in a major improvement in safety, the number of plane crashes has literally plummeted(sorry) after its introduction. But it hasn't eliminated crashes entirely - and every time there is a crash, Boeing/Airbus will ground their entire fleets to investigate what went wrong, if there is a perceived fault then thousands of planes have to be updated before they are safe to fly. Now bring that analogy over to cars - small, personal transportation vehicles, where profit margins really aren't that fantastic. Which manufacturer will risk releasing a product where they have to keep updating and maintaining the autopilot for the lifetime of the car, and for which they are liable every time it fails? Not to mention that planes are held to ridiculous safety standards, while cars are not(so far, I haven't heard an answer to what happens if a bird shits on the lidar and you don't clean it - it's an instrument with almost chirurgical precision which requires similar level of maintenance, and I know people who buy a car and then don't even wash it once in 3 years of owning it). I'm not saying we won't get there. I'm saying that I don't see any chance that this technology will be ready in 5-10 years, and if anyone is foolish enough to sell an "autonomous" car in the next 10 years, they will set back our progress massively because the car will fail, and there will be public and legislatory backlash.
  - Nomentatus 10y ago
    We get used to things that seem unsafe but are safe very quickly according to history. Elevators, for example. When I was young they generally had human operators, but now they're all autonomous, have been for a while, fail sometimes here and there but nobody seems hugely exercised about it.