Uber self-driving car kills pedestrian in first fatal autonomous crash

[dannno]

Ya I agree, it's amazing what you can see at night in the dark that a camera won't pickup, unless it has a flash..

 

[Danke]

Ya I agree, it's amazing what you can see at night in the dark that a camera won't pickup, unless it has a flash..

Actually, the system should see (detect) more in the dark than a human.

Uber and Waymo, which was spun off from Google, use lidar and radar technology, along with computer vision, to help guide the vehicle.

https://www.nytimes.com/interactive/2018/03/20/us/self-driving-uber-pedestrian-killed.html

 

[acptulsa]

Actually, the system should see (detect) more in the dark than a human.
https://www.nytimes.com/interactive/2018/03/20/us/self-driving-uber-pedestrian-killed.html

Sounds good in theory, doesn't it?

There are a few different types of technologies that are used in autonomous driving systems. Uber and Waymo, which was spun off from Google, use lidar and radar technology, along with computer vision, to help guide the vehicle.

A self-driving car’s sensors gather data on nearby objects, like their size and rate of speed. It categorizes the objects — as cyclists, pedestrians or other cars and objects — based on how they are likely to behave.

The thing is, the radar is liable to be optimized for spotting certain things like curbs, rather than spotting things moving into the path of the vehicle, as demonstrated by where the units are mounted in that diagram. And isn't optimal for spotting soft tissue like human clothing and flesh, which is pretty stealthy. Lidar requires a reflective thing to bounce laser beams off of, which black clothing is not, and I believe it sends pulses rather than a continual sweep. And as I mentioned, any camera which is optimized to make a picture that looks good is not optimized to show objects which are either underexposed or overexposed compared to the bulk of the surroundings--which means that specialized cameras need to be developed, as all existing cameras are designed to make pretty pictures.

Add in the fact that they don't seem interested in adding sound detectors and they certainly aren't stuffing IBM's Watson in the trunk of the vehicle, and I think it's safe to say these things are blind, deaf and dumb--especially at night.

 

[Brian4Liberty]

This is precisely what I was coming in here to say.

The ever fucking computers cannot intuit that there is a potential obstacle/incident about to happen.

They can only react to outside stimulus or pre-programming.

Motorcycle riding requires this sort of foresight, or you will end up dead.

So it's a safe bet you won't be signing up to test the first self-driving motorcycle?

I watched it about 10 times. I can't see her until she's in the headlights. The car should have probably been able to detect her because it has better sensory inputs. But I'm not sure this is a reason to scrap the technology.

Acptulsa pretty much covered it. The camera is very different from a human eye in the real world. There is still the potential that it was so dark that the person in the street was invisible, but probably not completely. We will never know for sure, as the back-up human eyes were on a cell phone at the time.

Are you out of your mind?

A human would not have been able to stop in time even if they were looking. Swerve and avoid? maybe but unlikely.

This is clearly the fault of the idiot pedestrian who was:

Are you high?

At a higher speed, swerving is usually the best method to avoid, but with enough warning, stopping is often possible.

And yes, it is a Darwin Award candidate who walks out on busy streets at night in front of traffic.

Doubtful. The car could have been relying on that very camera. The car's low beams were certainly aimed too low--polite to oncoming drivers, but not too useful. Most human eyes would have detected her in spite of that. Cameras adjust their aperture so they get the right amount of light in the largest part of their field if view. Human eyes do, too--but human eyes and brains have a much greater ability to see what is in the rest of the field of view--even if it's too bright or too dark--than any camera.

Regardless of what you see in the video, the "attendant" would have seen the victim in time to do something, had she been looking.

I'd have been able to change lanes safely and in time driving anything smaller than a tractor-trailer. And even in a semi I might have been able to spare her life, though probably not without damage to the vehicle.

And the lane was clear, too. That senseless death would have easily been avoided, had any human that didn't have night blindness been in that car behaving responsibly.

And no, Matt, I did not say the pedestrian demonstrated a lick of sense. I'm just saying that whether there was time to stop or not, there was all the time in the world to change to the left lane.

That about covers it. I'll bet by the end the human driver will end up with some repercussions. Odds are, the contract for being a test driver includes that they must be diligent and follow all laws the same as if they were driving themselves. And the driver facing camera is a witness to lack of attention.

 

[specsaregood]

That about covers it. I'll bet by the end the human driver will end up with some repercussions. Odds are, the contract for being a test driver includes that they must be diligent and follow all laws the same as if they were driving themselves. And the driver facing camera is a witness to lack of attention.

Yeah, I can't see how the driver could rightfully get out of some serious charges, it was clearly a case of distracted driving. No different than if it was a non-robot car. I doubt the law makes a distinction.

 

[acptulsa]

Yeah, I can't see how the driver could rightfully get out of some serious charges, it was clearly a case of distracted driving. No different than if it was a non-robot car. I doubt the law makes a distinction.

The driver's only possible defense is that the pedestrian was not in a crosswalk. Which should work in court, but seems awfully weak in humanitarian terms.

 

[Matt Collins]

I'd have been able to change lanes safely and in time driving anything smaller than a tractor-trailer. And even in a semi I might have been able to spare her life, though probably not without damage to the vehicle.

And the lane was clear, too. That senseless death would have easily been avoided, had any human that didn't have night blindness been in that car behaving responsibly.

Unlikely anyone's reaction time was enough to swerve in time.

 

[acptulsa]

Unlikely anyone's reaction time was enough to swerve in time.

Incorrect.

 

[dannno]

Unlikely anyone's reaction time was enough to swerve in time.

That is by looking at the video, but it is very possible the driver, had they been attentive, could have seen them in the shadows. You know how shitty cell phone video and pictures, and video in general is at night without a light source.

But we will never know, because they weren't looking and the situation would be difficult to recreate precisely.

 

[angelatc]

Regardless of what you see in the video, the "attendant" would have seen the victim in time to do something, had she been looking.

Do something, yes. But not enough time to brake and/or swerve entirely.

 

[acptulsa]

And I'm telling all of you I have changed lanes in less space than that in a five thousand pound 1971 Imperial Lebaron, riding on four bias ply tires, while going faster than that.

And I'm talking about the distance at which she becomes visible in the video, never mind the distance at which she became visible to the naked eye.

 

[specsaregood]

Do something, yes. But not enough time to brake and/or swerve entirely.

even a slight reaction might have prevented death though.

 

[angelatc]

Acptulsa pretty much covered it. The camera is very different from a human eye in the real world.

Self-driving cars use a lot more than a camera to "see."

 

[angelatc]

even a slight reaction might have prevented death though.

Human drivers hit people every day. So far, I suspect the robots have a much better track record than we do.

 

[acptulsa]

If anyone doubts the human eye can do better than a camera, try this: Wait until dark, gather three or four pretty girls and a camera, and go to where you have a nice view of a city skyline. Gather the girls together in the foreground with the lights in the buildings in the background, and take a picture.

The only way you'll reproduce what your eye sees in a picture is to either push the exposure to the point where you get movement blur, or take two pics--one with flash and one without--and combine them using photoshop or some such program.

Self-driving cars use a lot more than a camera to "see."

And in this case, those systems clearly failed. Or do you think the car was programmed to plow into obstacles in the road if they aren't in crosswalks?

Human drivers hit people every day. So far, I suspect the robots have a much better track record than we do.

Per hundred miles driven? Not even close. Computer cars' record as an accident rate is abysmal.

I'm not saying this technology is going nowhere. What I'm saying to you is, it ain't ready for prime time.

 

[dannno]

Do something, yes. But not enough time to brake and/or swerve entirely.

How do you know? If he was paying attention, he may have seen her in the shadows and slowed down like 5 seconds before the crash happened..

 

[dannno]

Human drivers hit people every day. So far, I suspect the robots have a much better track record than we do.

But there are only a few of them and millions of us.. I don't think their track record is all that great, but you would have to do an analysis of miles driven.

 

[Madison320]

Anyone here ever heard the phrase taught in Driver's Ed, "overdriving your headlights"? The premise is that you should be able to stop in time to avoid anything that becomes visible in your headlights.



https://www.ontario.ca/document/official-mto-drivers-handbook/driving-night-and-bad-weather

"Overdriving your headlights

You are overdriving your headlights when you go so fast that your stopping distance is farther than you can see with your headlights. This is a dangerous thing to do, because you may not give yourself enough room to make a safe stop. Reflective road signs can mislead you as well, making you believe you can see farther than you really can. This may cause you to over-drive your headlights if you are not careful "

 

[angelatc]

How do you know? ..

The same way everybody else knows that any and every human would certainly stopped in time, I suppose.

 

[tod evans]

The same way everybody else knows that any and every human would certainly stopped in time, I suppose.

1/2 the 'humans' I see are thumb-fucking their phones behind the wheel........