3 ms·

This was the original story: "After the Uber collision, the car continued traveling at 38 miles per hour, according to the Tempe police chief" https://www.bloo

by baking 9y ago

This was the original story: "After the Uber collision, the car continued traveling at 38 miles per hour, according to the Tempe police chief" https://www.bloomberg.com/news/articles/2018-03-21/for-self-driving-cars-seeing-everything-isn-t-always-enough https://www.bloomberg.com/news/articles/2018-03-21/for-self-... In other words, the car _never_ detected the pedistrian and never slowed down on its own. This has nothing to do with _when_ it saw her. It clearly didn't, yet both camera and LIDAR should have been able to.

ikeyany 9y ago
I feel bad for Waymo and all the guys taking the necessary safety precautions. This is going to set the industry back quite a bit. Thanks Uber, and RIP Elaine.
- InclinedPlane 9y ago
  I don't. Yes, I think Waymo is probably taking very nearly all the necessary safety precautions. But how would we know? It's simply a matter of trust, which is not good enough. And as we see here, the problem isn't Waymo the problem is every other fly-by-night automated vehicle operator out there. We should all know how the sausage gets made in this industry and it is absolutely not through a process of meticulous rigor and studious observance of industry best practices. We all know that software systems get made via mountains of short-cuts and compromises amidst an environment of half-assed development practices. The only way to fix this is to have regulations which actually enforce quality control. That means code audits, it means a lot more process and a lot more oversight, all of which is going to be a drag but which is necessary when developing systems on which so many lives could depend. If you look at the regulatory requirements for aviation software or even video poker machines they will really put into perspective how little is being done now with autonomous vehicles.
kalleboo 9y ago
This doesn't even make sense, we can see in the video that the "driver" does look up and react. Why didn't they hit the brakes?
- badthingfactory 9y ago
  At least one study been done to see how responsive a passive "driver" of an autonomous vehicle can be. The study concluded a person that is required to intervene with an autonomous vehicle is no better than a drunk person. It makes sense. If the car is able to get you from point A to point B 99.9% of the time without an issue, are you really going to be paying attention that one time when a person walks out in front of you and the car fails to respond appropriately? If the vehicle enables people to not pay attention, they aren't going to be attentive. I'd be willing to bet almost every driver that gets to ride around in these test vehicles thinks the job is awesome because they get paid without having to really do anything.
  - patentatt 9y ago
    Maybe this suggests that an effective use of current (i.e., 99.9%) autonomous driving implementations would be a captain and co-captain type of scenario, where the car only responds to commands that are the same from both drivers, like a logical AND between driver inputs (within some tolerance, and filtered, etc.). What if the promise of autonomous driving tech isn't to make human drivers obsolete, but to make human drivers nearly perfectly safe. What are the chances that an engaged human driver and an autonomous system would both, simultaneously give the same incorrect or dangerous driver input to a car? It's not as sexy as robot cars, but seems like a significant development over the current situation. Also, it seems like a viable way to test on public roads.
    - stordoff 9y ago
      What do you do in the case where the driver and the computer disagree? Say there is an object directly in front of the car - driver turns left to avoid, computer turns right to avoid. You have to take SOME action -- you can't do nothing otherwise you hit the obstacle. If you go with the driver, what is the point of the computer if it can't overrule bad decisions? That's equivalent to just letting the driver drive. If you go with the computer, what is the point of the driver? Only taking actions due to an AND of driver and computer inputs seems to only work if there is one correct course of action, which I doubt is the case in many situations. Maybe you could have a system where the computer only takes actions to deliberately prevent unsafe situations, and is conservative in doing so (i.e. it doesn't drive, per se, but brings the car to a safe stop or enforces a maximum speed), but that's a huge step down from the current goals.

by baking 9y ago

ikeyany 9y ago

I feel bad for Waymo and all the guys taking the necessary safety precautions. This is going to set the industry back quite a bit. Thanks Uber, and RIP Elaine.

InclinedPlane 9y ago
I don't. Yes, I think Waymo is probably taking very nearly all the necessary safety precautions. But how would we know? It's simply a matter of trust, which is not good enough. And as we see here, the problem isn't Waymo the problem is every other fly-by-night automated vehicle operator out there. We should all know how the sausage gets made in this industry and it is absolutely not through a process of meticulous rigor and studious observance of industry best practices. We all know that software systems get made via mountains of short-cuts and compromises amidst an environment of half-assed development practices. The only way to fix this is to have regulations which actually enforce quality control. That means code audits, it means a lot more process and a lot more oversight, all of which is going to be a drag but which is necessary when developing systems on which so many lives could depend. If you look at the regulatory requirements for aviation software or even video poker machines they will really put into perspective how little is being done now with autonomous vehicles.

kalleboo 9y ago

This doesn't even make sense, we can see in the video that the "driver" does look up and react. Why didn't they hit the brakes?

badthingfactory 9y ago
At least one study been done to see how responsive a passive "driver" of an autonomous vehicle can be. The study concluded a person that is required to intervene with an autonomous vehicle is no better than a drunk person. It makes sense. If the car is able to get you from point A to point B 99.9% of the time without an issue, are you really going to be paying attention that one time when a person walks out in front of you and the car fails to respond appropriately? If the vehicle enables people to not pay attention, they aren't going to be attentive. I'd be willing to bet almost every driver that gets to ride around in these test vehicles thinks the job is awesome because they get paid without having to really do anything.
- patentatt 9y ago
  Maybe this suggests that an effective use of current (i.e., 99.9%) autonomous driving implementations would be a captain and co-captain type of scenario, where the car only responds to commands that are the same from both drivers, like a logical AND between driver inputs (within some tolerance, and filtered, etc.). What if the promise of autonomous driving tech isn't to make human drivers obsolete, but to make human drivers nearly perfectly safe. What are the chances that an engaged human driver and an autonomous system would both, simultaneously give the same incorrect or dangerous driver input to a car? It's not as sexy as robot cars, but seems like a significant development over the current situation. Also, it seems like a viable way to test on public roads.
  - stordoff 9y ago
    What do you do in the case where the driver and the computer disagree? Say there is an object directly in front of the car - driver turns left to avoid, computer turns right to avoid. You have to take SOME action -- you can't do nothing otherwise you hit the obstacle. If you go with the driver, what is the point of the computer if it can't overrule bad decisions? That's equivalent to just letting the driver drive. If you go with the computer, what is the point of the driver? Only taking actions due to an AND of driver and computer inputs seems to only work if there is one correct course of action, which I doubt is the case in many situations. Maybe you could have a system where the computer only takes actions to deliberately prevent unsafe situations, and is conservative in doing so (i.e. it doesn't drive, per se, but brings the car to a safe stop or enforces a maximum speed), but that's a huge step down from the current goals.