Self-Driving Uber Car Kills Arizona Pedestrian (Phoenix: construction, income, costs)

[Cloudy Dayz]

Originally Posted by Bitey

My first takeaway is that those headlights either completely suck or they were severely misaligned if the car was out-driving their effective illumination distance at a mere 38 miles an hour. I think automotive designers and engineers need to spend a little less time worrying about how cool their headlights look and spend a little more time making sure they're actually effective.

Actually those headlights illuminated almost exactly the same as the low beam headlights on my Honda Civic. Which reminds me that I have to make getting some LED headlights a priority.

[cattastic]

A lot of people commenting here don't understand the basics of the technology at play.

The cars have sensors that don't depend on daylight in order to detect road hazards and the changing conditions of the road. The fact that the vehicle didn't even slow down means its sensors did not detect the person in the road. This is the fundamental problem, and why the issue is being taken so seriously.

Yes, the person should not have been jaywalking but there is a clear flaw in the technology that needs to be addressed before we can deem them entirely safe out in the road.

I actually just attended a conference run by a couple guys from MIT developing their own self-driving car and they went into a lay-persons explanation of all the different technologies being used on these vehicles. The cars can see in the infrared, they use lasers to make a 3-d map of the entire environment as the vehicle is driving. They go through hundreds of thousands of hours of testing to "teach" the computer how to identify and classify the world around it: the technology should be able to tell the difference between a person and a lamppost, for example.

In its testing phase, some of this technology is designed to actually get "smarter". The more encounters it has, the more accurate its reactions become by expanding its data set and increases its ability to predict what might happen next.

The fact that the car did not slow means it was not properly detecting the environment around it. It should have seen the pedestrian long before the headlights did. Which means there was some kind of technical failure.

I am very interested to see how the investigation pans out to identify the exact issue.

[Bitey]

Originally Posted by cattastic

A lot of people commenting here don't understand the basics of the technology at play.

The cars have sensors that don't depend on daylight in order to detect road hazards and the changing conditions of the road. The fact that the vehicle didn't even slow down means its sensors did not detect the person in the road. This is the fundamental problem, and why the issue is being taken so seriously.

Yes, the person should not have been jaywalking but there is a clear flaw in the technology that needs to be addressed before we can deem them entirely safe out in the road.

I actually just attended a conference run by a couple guys from MIT developing their own self-driving car and they went into a lay-persons explanation of all the different technologies being used on these vehicles. The cars can see in the infrared, they use lasers to make a 3-d map of the entire environment as the vehicle is driving. They go through hundreds of thousands of hours of testing to "teach" the computer how to identify and classify the world around it: the technology should be able to tell the difference between a person and a lamppost, for example.

In its testing phase, some of this technology is designed to actually get "smarter". The more encounters it has, the more accurate its reactions become by expanding its data set and increases its ability to predict what might happen next.

The fact that the car did not slow means it was not properly detecting the environment around it. It should have seen the pedestrian long before the headlights did. Which means there was some kind of technical failure.

I am very interested to see how the investigation pans out to identify the exact issue.

My speculation is that the computer didn't process the victim as a pedestrian because she was walking with a bike and thus presented a much different form factor than a human pedestrian alone when viewed perpendicular to the car's path of travel. If that's the case, the question is what DID the computer think the victim and her bike were? Another vehicle perhaps that it didn't recognize in time was swerving into its lane?

[cattastic]

Originally Posted by Bitey

My speculation is that the computer didn't process the victim as a pedestrian because she was walking with a bike and thus presented a much different form factor than a human pedestrian alone when viewed perpendicular to the car's path of travel. If that's the case, the question is what DID the computer think the victim and her bike were? Another vehicle perhaps that it didn't recognize in time was swerving into its lane?

I doubt it. Regardless of it's classification of the object, it should have detected that it was moving and its predictive technology is good enough that it should have been able to process that the object would continue moving into the car's path.

The sensors were working well enough to keep the car on the road and navigating, so I'm not sure we can speculate exactly what went wrong, as laypeople.

I am eagerly looking forward to more information on the investigation. Also I don't know which specific sensors and technology this particular car had.

I really wanted to email one of the guys that did the conference I attended because I figured they could make an educated guess about which component likely wasn't working properly.

[Bitey]

Originally Posted by cattastic

I doubt it. Regardless of it's classification of the object, it should have detected that it was moving and its predictive technology is good enough that it should have been able to process that the object would continue moving into the car's path.

The sensors were working well enough to keep the car on the road and navigating, so I'm not sure we can speculate exactly what went wrong, as laypeople.

I am eagerly looking forward to more information on the investigation. Also I don't know which specific sensors and technology this particular car had.

I really wanted to email one of the guys that did the conference I attended because I figured they could make an educated guess about which component likely wasn't working properly.

We can speculate all we want as laypeople as long as we understand our speculation is nothing more than conjecture... which is a synonym for speculation.

What's stopping you from emailing that guy you want to email? I'm interested in his educated guess too.

[cattastic]

I don't have his email!

The fact is, the LIDAR and SONAR is designed to detect moving objects. This is not an issue of "the car didn't see"--the car should have but something malfunctioned or went terribly, terribly wrong. This is inherently against the design of the car.

Human reaction time is very slow. Self driving cars should have a much faster and more accurate reaction time in events like this.

Something malfunctioned. That's not just conjecture; that's a fact if you have a basic understanding of the technology. WHAT and HOW and WHY are the reasons I am very curious about!


I don't have the guy's email! I am trying to look up his name.

[craig11152]

Here is a couple youtube videos of that same stretch of road at night. They certainly suggest the Uber camera is a bit deceiving with respect to visibility. They kind of dispell the "jumped out of the dark" notion.
The first one has no audio but the poster said the location is at about 33-36 second mark.

https://www.youtube.com/watch?time_c...&v=CRW0q8i3u6E




https://www.youtube.com/watch?v=1XOVxSCG8u0

[cattastic]

I don't know why we are focusing on how dark it was.

LIDAR and SONAR are equipped on these vehicles. The issue isn't that it was dark. The issue is that the LIDAR/SONAR did not act as it was designed to do.

Why?

Was the LIDAR turned off? Is it an issue with the software?

LIDAR and SONAR do not rely on visible light to detect the environment. That's the whole POINT of self driving cars!

I am very eager to see what the investigation finds, and who will be at fault: the company making the hardware, the company making the software, or Uber. This is a fascinating and unprecedented incident that will shape the future.

[lvmensch]

Originally Posted by cattastic

I don't know why we are focusing on how dark it was.

LIDAR and SONAR are equipped on these vehicles. The issue isn't that it was dark. The issue is that the LIDAR/SONAR did not act as it was designed to do.

Why?

Was the LIDAR turned off? Is it an issue with the software?

LIDAR and SONAR do not rely on visible light to detect the environment. That's the whole POINT of self driving cars!

I am very eager to see what the investigation finds, and who will be at fault: the company making the hardware, the company making the software, or Uber. This is a fascinating and unprecedented incident that will shape the future.

It is LIDAR and RADAR that should have easily identified the lady. The RADAR is crude and would just pick up something there but should have gotten a good echo to warn. The LIDAR should have clearly picked her up as something good sized moving in a dangerous direction. A couple of scans and it should have had her clearly targeted. This is big basic stuff. That is why the event is so troubling.

Sonar is generally short range stuff a k- cars in the blind spot or parking. Not detecting at range.

[craig11152]

Originally Posted by cattastic

I don't know why we are focusing on how dark it was.

I am speaking to it in part because we have two parallel issues. One, you are beating to death, is that the car malfunctioned.
But there is a parallel discussion that suggests the woman would have been killed by a human driven car because in the case of a human driven car the accident was unavoidable.Ergo driverless cars are safe. That is simply not so clear. Why does it matter? Because it speaks to whether these cars are now, or ever will be, safer overall than human driven cars. In order to make that claim they need to kill less people than human drivers do. In the case of pedestrians that is one every 688 million miles. Overall that is 1 every 100 million miles. So far between Uber and Tesla they are not meeting that number.