9 ms·

> When they became fully visible, our vehicle applied heavy braking but was not able to avoid the collision. The "fully visible" part made me thinks humans can

by itake 3y ago

> When they became fully visible, our vehicle applied heavy braking but was not able to avoid the collision. The "fully visible" part made me thinks humans can identify a bicycle and rider with just partial visibility, which means they would stop faster than the car.

roenxi 3y ago
Based on what? Humans run in to cyclists too and although I have know idea about Waymo's tech, in a research setting computers are better than humans at figuring out what objects are based on visual data.
- al_borland 3y ago
  If computers are so good at this, why am I (a human) asked to identify bicycles and motorcycles to train these computers when confronted with a captcha? A human needs far less training and can identify a person on a bike without ever seeing an actual bike. They can also infer what is likely to happen based on a lot of context clues. Even if an event doesn’t seem likely, I can still prepare. For example, if I see kids playing basketball in a driveway, I’m going to slow down, incase the ball bounces into the street and a kid takes off after it without looking. I don’t need to wait for the kid to be in the street to start reacting and prepare for different possibilities.
  - ganieldackle 3y ago
    A computer doesn't need a multi million dollar marketing campaign convincing it not to text while driving or drive drunk. >I don’t need to wait for the kid to be in the street to start reacting and prepare for different possibilities. Neither does a computer.
    - patrick451 3y ago
      > A computer doesn't need a multi million dollar marketing campaign convincing it not to text while driving or drive drunk. A human doesn't get random bugs every week from OTA updates. The human mind is not a viable target for hackers. The risks of human drivers are individual. The risks of computer drivers are systemic and massive. One bad bug or hack could kill millions if self driving cars become ubiquitous.
      - ganieldackle 3y ago
        >A human doesn't get random bugs every week from OTA updates. What bugs are you referencing exactly? Humans have something called emotions that are far more variable than bugs in a safety critical system and you're exposed to thousands of different flavors every time you get on the road. >The human mind is not a viable target for hackers. But it is, and not even sophisticated ones at that. One lane change and a middle finger from a fellow driver can induce a murderous road rage. >The risks of computer drivers are systemic and massive. Hypothesis. Reality: 45,000 people die every year in the US from human drivers. That is systemic and massive.
        patrick451 3y ago
        You still don't get. All human problems are individual. Emotions in one person have zero impact on emotions in a driver two state away. Software problems are systemic. They change the entire system at once. Your "just hypothesis" dismissal would carry a lot more weight if silicon valley hadn't been cranking out bug filled garbage for the last 20 years in their move fast and break things frenzy. Kyle Vogt was (in)famous for vaunting that culture at cruise so there's no argument AVs have somehow escaped the silicon valley rot. The tech industry need to stay far, far away from anything safety related
      - scarmig 3y ago
        > One bad bug or hack could kill millions if self driving cars become ubiquitous. Every year, human drivers kill around 1.2 million people every single year (https://www.who.int/news-room/fact-sheets/detail/road-traffic-injuries https://www.who.int/news-room/fact-sheets/detail/road-traffi...).
      - jdietrich 3y ago
        Humans have a bunch of unpatchable bugs in safety-critical systems. https://en.m.wikipedia.org/wiki/Inattentional_blindness https://en.m.wikipedia.org/wiki/Inattentional_blindness https://en.m.wikipedia.org/wiki/Saccadic_masking https://en.m.wikipedia.org/wiki/Saccadic_masking
      - stqism 3y ago
  - aetch 3y ago
    The computers don’t need you to do captchas anymore. The captchas are just solved problems being used to slow down bots and spammers these days.
    - al_borland 3y ago
      Are they though? https://news.ycombinator.com/item?id=37160744 https://news.ycombinator.com/item?id=37160744
      - roenxi 3y ago
        You're linking to discussion of an article that explains bots are solving CAPTCHA and reCAPTCHA faster and more accurately than humans. I don't understand what you are trying to say, but I don't think it aligns with the evidence you're providing.
        al_borland 3y ago
        You mentioned… >The captchas are just solved problems being used to slow down bots and spammers these days. If bot are solving these problems faster and more accurately than humans, it’s not really slowing down the bots and spammers anymore. That is also a solved problem.
        roenxi 3y ago
        Well, if it is solving CAPTCHA better and faster than humans, that is a useful signal. Maybe they monitor how often an IP address figures something out and if it gets too many right they +1 a suspicion counter. I dunno. Implementing the CAPTCHA solving bot also takes time and expertise; it'd filter out some low-ability spammers.
  - data-ottawa 3y ago
    The captcha example isn't exactly fair, the images shown can be newly generatrd, repeated, or used as control images, and even if they are unclassifiable then you're not seeing the successes. Without knowing the whole system there's too much hidden bias to claim the computer is more of less accurate than a commuter.
    - tiahura 3y ago
      No doubt all of your excuses are at least partially true, however, at this point, literally after almost two decades of billions of people training these things to see bikes, it still needs a lot more work.
      - jhgg 3y ago
        Recaptcha has not been used to train any neural network to detect anything in a very long time. In fact, visual challenges like this are a mostly solved problem, and the real magic is the classification that happens behind the scenes.
        al_borland 3y ago
        How many years or decades did we spend answering them before it got to where it today? I’m all for optimism, but these problems are clearly much harder than the futurists would like the public to believe. Elon said Teslas would be driving across the country to pick up their owners by 2018. 6 years later and what does it do, drive across a parking lot? https://twitter.com/elonmusk/status/686279251293777920 https://twitter.com/elonmusk/status/686279251293777920 Trains are also a solved problem, and seem like a much better solution to the problem. With rails, a vast majority of the complexity around driving goes away. They also remove traffic instead of keeping it the same or making it worse. Less traffic would likely mean fewer accidents, even if humans are still the ones driving. And batching transit together, with dozens of people in one vehicle is going to be far more energy efficient than everyone running their own vehicle. It also solves the issue of drunk driving, distracted driving, and bad drivers in general, as everyone is just a passenger. So much money, time, talent, and energy has gone into this self driving stuff, and it seems like a massive misallocation of resources. They’re solving the wrong problem.
  - shiroiuma 3y ago
    >if I see kids playing basketball in a driveway, I’m going to slow down, incase the ball bounces into the street and a kid takes off after it without looking That's because you're a thoughtful, conscientious driver. A very large fraction of drivers are not, so society winds up with lots of dead kids. Humans are terrible drivers: some of them are OK, but too many are very bad, leading to catastrophic results, and because of this human drivers need to be eliminated. Alternatively, the society could become far stricter about which humans are allowed to drive, but that isn't feasible when the society has intentionally made it so that it's effectively impossible to survive without owning and driving a car.
- bunabhucan 3y ago
  Seeing a bike and inferring the vector of a bike are very different. A kid or tweaker on a bike needs more frequent updates than someone commuting or training.
seanmcdirmid 3y ago
> The "fully visible" part made me thinks humans can identify a bicycle and rider with just partial visibility, which means they would stop faster than the car. From my experience as a driver and a bicyclist, I guarantee you that is not true, most driver won’t see you unless you are >60% visible.
r0m4n0 3y ago
Last year I caught this footage of a driver that hit a cyclist and they were fully visible. The police determined the driver technically wasn’t at fault because they had the right of way but humans drive incredibly unsafe. https://youtu.be/uycFSrUI700?si=DPgNc_B5CNDctDyt https://youtu.be/uycFSrUI700?si=DPgNc_B5CNDctDyt 1) cyclist assumed they were safe to cross as there was a pedestrian 2) the driver didn’t slow down for the pedestrian in the road at all 3) driver was probably speeding 4) driver didn’t brake until very late, probably distracted 5) cyclist was 15 years old so probably not using enough caution in general I honestly think a self driving car would have been going slow for the crossing pedestrian and this collision would have never happened (but I guess we can all just guess). The question is, what’s better, occasional software glitches or human stupidity? I guess time and data will tell.
- fallingknife 3y ago
  Not sure what the relevance of the pedestrian is because he far outside the path of travel for the car at the time he hit the bike.
  - cozzyd 3y ago
    most states require you to stop as long as a pedestrian is in the cross-walk (one of the many traffic rules that most drivers disregard).
    - seanmcdirmid 3y ago
      Some states have a half rule. Or a rule that as long as the pedestrian hasn’t passed your car yet, you need to stop. I’m not even clear what my state (Washington) follows ATM: > (1) The operator of an approaching vehicle shall stop and remain stopped to allow a pedestrian, bicycle, or personal delivery device to cross the roadway within an unmarked or marked crosswalk when the pedestrian, bicycle, or personal delivery device is upon or within one lane of the half of the roadway upon which the vehicle is traveling or onto which it is turning. For purposes of this section “half of the roadway” means all traffic lanes carrying traffic in one direction of travel, and includes the entire width of a one-way roadway.
minhazm 3y ago
Even if a human sees the bike slightly sooner, the reaction time is so much slower that it's either a wash or the self driving car is likely still faster. Median braking reaction time for humans is ~500ms[1]. [1] https://pubmed.ncbi.nlm.nih.gov/26376036/ https://pubmed.ncbi.nlm.nih.gov/26376036/
- nomel 3y ago
  I can't read that article, but I suspect they're not including all the required foot movement there. This study [1] shows ~2 seconds from gas lift to max brake force, when using right foot. Eliminating the foot movement is why I left foot brake. I feel like I have some superpower with how fast I can react, compared to others. After all, we were taught to brake with our right foot out of tradition, not data! [1] https://copradar.com/redlight/factors/IEA2000_ABS51.pdf https://copradar.com/redlight/factors/IEA2000_ABS51.pdf

by itake 3y ago

roenxi 3y ago

Based on what? Humans run in to cyclists too and although I have know idea about Waymo's tech, in a research setting computers are better than humans at figuring out what objects are based on visual data.

al_borland 3y ago
If computers are so good at this, why am I (a human) asked to identify bicycles and motorcycles to train these computers when confronted with a captcha? A human needs far less training and can identify a person on a bike without ever seeing an actual bike. They can also infer what is likely to happen based on a lot of context clues. Even if an event doesn’t seem likely, I can still prepare. For example, if I see kids playing basketball in a driveway, I’m going to slow down, incase the ball bounces into the street and a kid takes off after it without looking. I don’t need to wait for the kid to be in the street to start reacting and prepare for different possibilities.
- ganieldackle 3y ago
  A computer doesn't need a multi million dollar marketing campaign convincing it not to text while driving or drive drunk. >I don’t need to wait for the kid to be in the street to start reacting and prepare for different possibilities. Neither does a computer.
  - patrick451 3y ago
    > A computer doesn't need a multi million dollar marketing campaign convincing it not to text while driving or drive drunk. A human doesn't get random bugs every week from OTA updates. The human mind is not a viable target for hackers. The risks of human drivers are individual. The risks of computer drivers are systemic and massive. One bad bug or hack could kill millions if self driving cars become ubiquitous.
    - ganieldackle 3y ago
      >A human doesn't get random bugs every week from OTA updates. What bugs are you referencing exactly? Humans have something called emotions that are far more variable than bugs in a safety critical system and you're exposed to thousands of different flavors every time you get on the road. >The human mind is not a viable target for hackers. But it is, and not even sophisticated ones at that. One lane change and a middle finger from a fellow driver can induce a murderous road rage. >The risks of computer drivers are systemic and massive. Hypothesis. Reality: 45,000 people die every year in the US from human drivers. That is systemic and massive.
      - patrick451 3y ago
        You still don't get. All human problems are individual. Emotions in one person have zero impact on emotions in a driver two state away. Software problems are systemic. They change the entire system at once. Your "just hypothesis" dismissal would carry a lot more weight if silicon valley hadn't been cranking out bug filled garbage for the last 20 years in their move fast and break things frenzy. Kyle Vogt was (in)famous for vaunting that culture at cruise so there's no argument AVs have somehow escaped the silicon valley rot. The tech industry need to stay far, far away from anything safety related
    - scarmig 3y ago
      > One bad bug or hack could kill millions if self driving cars become ubiquitous. Every year, human drivers kill around 1.2 million people every single year (https://www.who.int/news-room/fact-sheets/detail/road-traffic-injuries https://www.who.int/news-room/fact-sheets/detail/road-traffi...).
    - jdietrich 3y ago
      Humans have a bunch of unpatchable bugs in safety-critical systems. https://en.m.wikipedia.org/wiki/Inattentional_blindness https://en.m.wikipedia.org/wiki/Inattentional_blindness https://en.m.wikipedia.org/wiki/Saccadic_masking https://en.m.wikipedia.org/wiki/Saccadic_masking
    - stqism 3y ago
- aetch 3y ago
  The computers don’t need you to do captchas anymore. The captchas are just solved problems being used to slow down bots and spammers these days.
  - al_borland 3y ago
    Are they though? https://news.ycombinator.com/item?id=37160744 https://news.ycombinator.com/item?id=37160744
    - roenxi 3y ago
      You're linking to discussion of an article that explains bots are solving CAPTCHA and reCAPTCHA faster and more accurately than humans. I don't understand what you are trying to say, but I don't think it aligns with the evidence you're providing.
      - al_borland 3y ago
        You mentioned… >The captchas are just solved problems being used to slow down bots and spammers these days. If bot are solving these problems faster and more accurately than humans, it’s not really slowing down the bots and spammers anymore. That is also a solved problem.
        roenxi 3y ago
        Well, if it is solving CAPTCHA better and faster than humans, that is a useful signal. Maybe they monitor how often an IP address figures something out and if it gets too many right they +1 a suspicion counter. I dunno. Implementing the CAPTCHA solving bot also takes time and expertise; it'd filter out some low-ability spammers.
- data-ottawa 3y ago
  The captcha example isn't exactly fair, the images shown can be newly generatrd, repeated, or used as control images, and even if they are unclassifiable then you're not seeing the successes. Without knowing the whole system there's too much hidden bias to claim the computer is more of less accurate than a commuter.
  - tiahura 3y ago
    No doubt all of your excuses are at least partially true, however, at this point, literally after almost two decades of billions of people training these things to see bikes, it still needs a lot more work.
    - jhgg 3y ago
      Recaptcha has not been used to train any neural network to detect anything in a very long time. In fact, visual challenges like this are a mostly solved problem, and the real magic is the classification that happens behind the scenes.
      - al_borland 3y ago
        How many years or decades did we spend answering them before it got to where it today? I’m all for optimism, but these problems are clearly much harder than the futurists would like the public to believe. Elon said Teslas would be driving across the country to pick up their owners by 2018. 6 years later and what does it do, drive across a parking lot? https://twitter.com/elonmusk/status/686279251293777920 https://twitter.com/elonmusk/status/686279251293777920 Trains are also a solved problem, and seem like a much better solution to the problem. With rails, a vast majority of the complexity around driving goes away. They also remove traffic instead of keeping it the same or making it worse. Less traffic would likely mean fewer accidents, even if humans are still the ones driving. And batching transit together, with dozens of people in one vehicle is going to be far more energy efficient than everyone running their own vehicle. It also solves the issue of drunk driving, distracted driving, and bad drivers in general, as everyone is just a passenger. So much money, time, talent, and energy has gone into this self driving stuff, and it seems like a massive misallocation of resources. They’re solving the wrong problem.
- shiroiuma 3y ago
  >if I see kids playing basketball in a driveway, I’m going to slow down, incase the ball bounces into the street and a kid takes off after it without looking That's because you're a thoughtful, conscientious driver. A very large fraction of drivers are not, so society winds up with lots of dead kids. Humans are terrible drivers: some of them are OK, but too many are very bad, leading to catastrophic results, and because of this human drivers need to be eliminated. Alternatively, the society could become far stricter about which humans are allowed to drive, but that isn't feasible when the society has intentionally made it so that it's effectively impossible to survive without owning and driving a car.
bunabhucan 3y ago
Seeing a bike and inferring the vector of a bike are very different. A kid or tweaker on a bike needs more frequent updates than someone commuting or training.

seanmcdirmid 3y ago

> The "fully visible" part made me thinks humans can identify a bicycle and rider with just partial visibility, which means they would stop faster than the car. From my experience as a driver and a bicyclist, I guarantee you that is not true, most driver won’t see you unless you are >60% visible.

r0m4n0 3y ago

Last year I caught this footage of a driver that hit a cyclist and they were fully visible. The police determined the driver technically wasn’t at fault because they had the right of way but humans drive incredibly unsafe. https://youtu.be/uycFSrUI700?si=DPgNc_B5CNDctDyt https://youtu.be/uycFSrUI700?si=DPgNc_B5CNDctDyt 1) cyclist assumed they were safe to cross as there was a pedestrian 2) the driver didn’t slow down for the pedestrian in the road at all 3) driver was probably speeding 4) driver didn’t brake until very late, probably distracted 5) cyclist was 15 years old so probably not using enough caution in general I honestly think a self driving car would have been going slow for the crossing pedestrian and this collision would have never happened (but I guess we can all just guess). The question is, what’s better, occasional software glitches or human stupidity? I guess time and data will tell.

fallingknife 3y ago
Not sure what the relevance of the pedestrian is because he far outside the path of travel for the car at the time he hit the bike.
- cozzyd 3y ago
  most states require you to stop as long as a pedestrian is in the cross-walk (one of the many traffic rules that most drivers disregard).
  - seanmcdirmid 3y ago
    Some states have a half rule. Or a rule that as long as the pedestrian hasn’t passed your car yet, you need to stop. I’m not even clear what my state (Washington) follows ATM: > (1) The operator of an approaching vehicle shall stop and remain stopped to allow a pedestrian, bicycle, or personal delivery device to cross the roadway within an unmarked or marked crosswalk when the pedestrian, bicycle, or personal delivery device is upon or within one lane of the half of the roadway upon which the vehicle is traveling or onto which it is turning. For purposes of this section “half of the roadway” means all traffic lanes carrying traffic in one direction of travel, and includes the entire width of a one-way roadway.

minhazm 3y ago

Even if a human sees the bike slightly sooner, the reaction time is so much slower that it's either a wash or the self driving car is likely still faster. Median braking reaction time for humans is ~500ms[1]. [1] https://pubmed.ncbi.nlm.nih.gov/26376036/ https://pubmed.ncbi.nlm.nih.gov/26376036/

nomel 3y ago
I can't read that article, but I suspect they're not including all the required foot movement there. This study [1] shows ~2 seconds from gas lift to max brake force, when using right foot. Eliminating the foot movement is why I left foot brake. I feel like I have some superpower with how fast I can react, compared to others. After all, we were taught to brake with our right foot out of tradition, not data! [1] https://copradar.com/redlight/factors/IEA2000_ABS51.pdf https://copradar.com/redlight/factors/IEA2000_ABS51.pdf