Uber

[Schlich]

I refuse to watch the video because I enjoy sleeping, but sometimes slowing down isn't the best approach to avoid a moving object. Sometimes it's better to speed up before it gets in your way.

it was at night, extremely dark, and the pedestrian very suddenly appears in the path of the headlights approximately 0.5 seconds before impact.

[Donnie Ebert]

I refuse to watch the video because I enjoy sleeping, but sometimes slowing down isn't the best approach to avoid a moving object. Sometimes it's better to speed up before it gets in your way.

it was at night, extremely dark, and the pedestrian very suddenly appears in the path of the headlights approximately 0.5 seconds before impact.

Based on what a camera can see? The eye can see much better at night than the quality of that video. A human likely would've seen something, perhaps not in time to stop or react, but likely something if they were paying attention.

Plus, the Uber was in the right lane, with the pedestrian/bike coming across traffic from the left, so the pedestrian/bike didn't just hop off a curb from behind a bush or something and right in front of the car. The sensors and software should've detected this, otherwise they should go back to testing in a controlled environment.

[Schlich]

I refuse to watch the video because I enjoy sleeping, but sometimes slowing down isn't the best approach to avoid a moving object. Sometimes it's better to speed up before it gets in your way.

it was at night, extremely dark, and the pedestrian very suddenly appears in the path of the headlights approximately 0.5 seconds before impact.

Based on what a camera can see? The eye can see much better at night than the quality of that video. A human likely would've seen something, perhaps not in time to stop or react, but likely something if they were paying attention.

Plus, the Uber was in the right lane, with the pedestrian/bike coming across traffic from the left, so the pedestrian/bike didn't just hop off a curb from behind a bush or something and right in front of the car. The sensors and software should've detected this, otherwise they should go back to testing in a controlled environment.

I think your definition of "likely" and "should have" is a lot different than mine. What percent of drivers do you think could have avoided collision? 50%? More? Less?

Plus I generally can't see things out of the range of my headlights. I'm not sure how human vision vs video footage is that different in that regard. I've almost hit several jaywalkers or bikers at night, paying full attention.

[Schlich]

Also, software and hardware failures are to be expected and we need to get over it. The only benchmark should be its performance vs human drivers, period. And that's on average, not micro-dissecting individual cases.

[Joe Shlabotnik]

Also, software and hardware failures are to be expected and we need to get over it. The only benchmark should be its performance vs human drivers, period. And that's on average, not micro-dissecting individual cases.

Software and hadware failures are to be expected. That is why driverless cars are a bad idea. I write software, that is why I think it is a bad idea. I also know what can happen to a compromised software system. That is why driverless cars are a bad idea.

Driverless cars are a bad idea.

[Schlich]

That's unbelievably, stupidly short sighted. 40,000 people in the US alone died from automobile accidents last year and you don't care. OK.

[Schlich]

Let's just not drive cars at all since the brakes fail every now and then

[Arthur Dent]

I think your definition of "likely" and "should have" is a lot different than mine. What percent of drivers do you think could have avoided collision? 50%? More? Less?

Plus I generally can't see things out of the range of my headlights. I'm not sure how human vision vs video footage is that different in that regard. I've almost hit several jaywalkers or bikers at night, paying full attention.

I don’t think there’s enough information from that video alone to make confident judgement about what’s possible, but it’s definitely true that human vision is vastly better at seeing contrast in scenes with large variations in lighting than raw camera footage does.

Try this experiment: take your phone/camera into a dark room like a bathroom and shine a flashlight on some detail on the wall. The camera will show either the unlighted surroundings with the flashlight spot completely washed out or only the lighted area with the background too dark to see, but not both at the same time. Look up from the screen with your eyes, and there’s no problem.

These problems have technical solutions, the car should have senses humans don’t etc, but the video doesn’t actually establish that the wreck was unavoidable. Further, the driver is clearly checked out as almost everyone will be if they can count on their driverless system to handle everything but rare emergencies, and this is one of the key problem with the technology for the moment and probably the near to medium term.

[Schlich]

I understand the point that was being made about video and contrast, but practically speaking, I have had several instances where I was paying full attention to the road and had a jaywalker appear in my sight at the very last minute as I whizzed by. The number of times where I've seen a jaywalker outside of my headlight path is probably less times than those instances. Can I say with 100% confidence that the operator would not have seen the pedestrian if (s)he was paying full attention? No, sorry for being hypberbolic i guess, but I'm extremely tired of the double standards we're holding these systems to.

agreed on all your last points.

[Arthur Dent]

Also, software and hardware failures are to be expected and we need to get over it. The only benchmark should be its performance vs human drivers, period. And that's on average, not micro-dissecting individual cases.

Software and hadware failures are to be expected. That is why driverless cars are a bad idea. I write software, that is why I think it is a bad idea. I also know what can happen to a compromised software system. That is why driverless cars are a bad idea.

Driverless cars are a bad idea.

Counterpoint: Driverless cars are a great idea because human drivers are a deadly and horrendous disaster. On the other hand, making this idea a widespread reality is not possible with current technology and will need to be approached with much greater care and oversight than goes into typical commercial software development. On the plus side, the automotive industry is already used to this, but on the minus side much of the safety standards are often a formal looking cover for a much more chaotic and uncontrolled reality.