After Fatal Uber Incident, Nvidia Defers Autonomous Car Testing On Public Roads

Status
Not open for further replies.

TJ Hooker

Titan
Ambassador
I'd be curious to know how this accident would figure into an accidents per km driven metric for autonomous vehicles, and then compare it to the same metric for non-autonomous vehicles. I have a suspicion that the former would still be lower than the latter...

Edit: km driven on public roads that is.
 

USAFRet

Titan
Moderator


For that, you'd have to slice off all the km driven in absolute perfect test conditions.
Then compare.

Eventually they'll be better than the overall human population.
And eventually after that, "better" once you discount all the drunk drivers.
 
After reviewing the released footage, which doesn't show any graphic carnage and doesn't need to for one to get an idea how the accident happened, it's reasonable to suspect that a human driver would have run the woman over much the same as the autonomous vehicle did. The woman was nowhere near a legal road crossing, wearing black, walking on a dark, poorly lit portion of the roadway near a vehicle divergence point, coming out of a group of trees in the median, where you wouldn't normally expect the common jaywalker to be, and pushing a bicycle which may have played a role in confusing the vehicles sensors. Also, it isn't as if the woman who had a history of vagrancy couldn't see the car coming before she stepped into the street and started crossing slowly. Some onus of responsibility needs to be put on the person who walked in front of the oncoming vehicle. There is and always will be a limit to the ability of a computer in preventing human deaths.
 

USAFRet

Titan
Moderator


Well, they're supposed to be "better" than human.

That could easily have been a kid chasing a ball.

And...https://www.zerohedge.com/news/2018-03-20/tempe-police-say-no-fault-uber-fatal-crash
"The car, which was traveling at 38 mph in a 35 mph"
Speeding?


(and the womans possible but not verified "history of vagrancy" has nothing to do with this)
 
I agree, these computer controlled vehicles are supposed to better than humans, but they are still subject to the realities of physics and other impossible to predict realities. A 100% perfect record is an impossibility. A significant reduction in human induced errors, however, is both a very good and seemingly achievable goal. I suspect there is good correlation between the statistics for motor vehicle safety and aircraft safety (which there are plenty of good numbers for), and if we take as an example, Boeing's statistics (again, plenty of studies seem to support this), you would see that in about 80% of aircraft incidents, the fault lies with the human in part or in full, not the aircraft, with the other 20% being failure of the aircraft or outside influences such as weather. I have a suspicion that the numbers for autonomous vehicle safety vs human controlled vehicle is going to swing widely in favor of the more automated systems, but first they have to get the measures to be apples to apples, as you said, so we're not giving unfair advantage to the level 3, 4, and 5 vehicles operating on test tracks or perfect weather conditions. My money is still on the autonomous vehicles, at least after they've been through the proper development work.

If a child chasing a ball runs out from behind sensor obscuring cover, no existing system can currently resolve the situation in a favorable manner for all involved. In the future, that may change. In the case of your example, the onus is on the parent or guardian letting the child run into the street in the first place, not the computer for failing to stop a moving vehicle in less time or with less information than is required. It's a tragedy whenever something like this happens, but it doesn't change the fact that a lapse of human judgement or control was involved in the incident. Maybe we should blame Sesame Street for not teaching children to look both ways before crossing the street, or perhaps the responsible adult was too busy texting to watch the child?

Hopefully the cause of the failure of the vehicle in this case to identify the woman and take correct avoidance measures can be found and a permanent solution enacted.

You can be sure that somebody would be issued a citation by the police department, and most articles would likely point this out, if the car was found to be speeding. There has been a case of a Google car being pulled over without a citation being issued for impeding traffic by driving at 24 MPH in a 35 MPH zone. However, Google has admitted that their cars are designed to exceed the speed limit when it's safer to do so, due to traffic conditions. Hard to say what Uber's cars are allowed to do in regards breaking speed limits without Uber commenting on that. Ultimately though, the news article you linked to, which I did read, claims a speed limit for the road which I assert it is factually incorrect, and can even be verified through Google maps street view. I also raise you a couple more articles, with the relevant paragraphs in quotes:

From the New York Times article, "How a Self-Driving Uber Killed a Pedestrian in Arizona"
The vehicle was going about 40 miles an hour on a street with a 45-mile-an-hour speed limit when it struck Ms. Herzberg, 49, who was walking her bicycle across the street, according to the Tempe police.
From the Reuters article, "Homeless Arizona woman killed by Uber self-driving SUV was 'like everyone's aunt'"
TEMPE, Ariz. (Reuters) - The Arizona pedestrian killed by an Uber self-driving vehicle was a homeless woman close to getting off the streets, her friends said, describing her as a fighter who took care of those around her.
I think we could argue, if you really wanted to, of the relevancy of both a person's financial status and psychological well-being, in how individuals interact with the places in which they find themselves. It doesn't tend to be the well dressed and well off I see breaking laws so they can exercise what for most appearances must be the worst possible decisions about where and how to cross the road.
 

jpe1701

Honorable


You had me until you started arguing that Her vagrancy had anything to do with it. Do you know the area? How do you know the status of her "psychological well-being"? And who do you see doing crimes? You think people that dress well don't commit crimes or have mental illness? Maybe they change clothes for you.
 
If the woman wasn't a vagrant, or homeless, she probably wouldn't have been crossing the street the way she was, at the time of day she did, so you could make the case that being homeless had everything to do with her death. She probably would be alive today had she not been homeless, so how dare you not think her homelessness had a factor to play in her death? Don't blame my dissatisfaction for her state in life to be some sort of disapproval or disdain for her as a person. You have limited means here to glean just how much compassion I have for people.

You question my knowledge of the area? I question yours as well. I didn't take a stance on her mental well-being, but I do think it's significant and called it into question. Does the average person cross the street in front of a moving automobile, in the dark, without concern for their own safety? I'm not going to discount her mental health, no matter what it was, just because she may be without a consistent place of residence. People in dire situations often make decisions that they may not otherwise, decisions which are often times not the best for anybody involved.

The moment she stepped into the road outside of a crosswalk, she broke the law. The laws covering jaywalking are there for the benefit of both the motorist and the pedestrian. Well dressed or not, she stepped in front of a moving vehicle, whether aware of that fact or not, and paid a sad, fatal consequence.

This is not an argument for or against the dignity or humanity of her person, that just happened to be in the situation she was in. The issue of homelessness is a discussion that is outside of the bounds of this article. But, it does intersect when you look at how homeless tend to behave in contrast to how people who aren't as concerned with food or shelter tend to behave.

I live in a suburb of an area with a large vagrant population. Arguing that they are in any way model citizens or subject to special consideration above the laws that are in place to protect citizens is not going to give you any traction with me, but it's good to keep the conversation going.

If we are going to progress beyond letting humans make errors while driving, maybe there is progress we can make in regards pedestrians making errors as well? Do we fault trains when they run over people that are walking on the tracks? Do people really think that no matter what a pedestrian does does on a road, they should be excused and the motorist at fault?
 

Ninjawithagun

Distinguished
Aug 28, 2007
747
16
19,165


You are absolutely correct on all points - well said! I have been discussing this very incident with coworkers and we all agree that Uber is not at fault here. Anybody would have hit this lady in almost the same way. We may have gotten lucky and hit the brakes or swerved at the last second. But in the end, I don't think it was unavoidable. If you watch the video, it's fairly conclusive that the pedestrian was not visible until the last second prior to being hit. It's a sad thing that someone died, but blame can't be blind either. If I were the judge, I would rule on the side of Uber due to so many contributing factors on behalf of the pedestrian.

 

Ninjawithagun

Distinguished
Aug 28, 2007
747
16
19,165


Keep your comments to yourself. No politics in the forums please - thank you!

 

Brian_R170

Honorable
Jun 24, 2014
288
2
10,785
I saw the footage and I am skeptical. The woman wasn't walking fast. She didn't jump out in front of the car. It's really hard to believe there isn't more contrast between the woman and the shadows when the car got that close with headlights on. The video captured and processed by the the car really should be capable of capturing much more detail in low light conditions than the released video shows. I would be very surprised if it doesn't, but if it doesn't, then I would certainly call the system defective.

Also, there are non-autonomous vehicles with collision avoidance systems that don't rely on video, so even if you believe that a human driver couldn't have avoided it, it is possible, or even likely, that at least some non-autonomous vehicles could have.
 

jpe1701

Honorable


Keep your comments to yourself. What does discussing the womans homelessness have to do with politics? I was mistaken about pinkdragons post because I thought he was disparaging the womans situation and it kind of sounded like he meant she deserved it because she was a vagrant. He made his point. I happen to agree on everything else you both have said.
Ps- I know absolutely nothing of the area, my point was there could have been a reason to cross there or maybe there was a business nearby where a well dressed person could have been hit. I didn't mean to come off as a jerk just provoking thought.
 

zodiacfml

Distinguished
Oct 2, 2008
1,228
26
19,310
Seeing the footage, there truly is a problem with its system. Somethign must have failed but it usually has redundancies.
More likely, the system was not able to detect the person and the bicycle. The driver behind the wheel seems to be looking at the screen without warnings.
 

hannibal

Distinguished
Human driver can easily hit a running child... so can AI.
There is almost zero time to react to that. Ai can be quicker, Human can predict problems even before They happens, but They seldom do that and that is why there Are so Many accident with cars and running childrens.

Human can know that there live children in this and that house and be more alert in close of those houses, but even that is not enough some times.
All in all it is still a matter who is responsible when these things happens. The people in the car, the car manufacturer, the autodrive hardsware manufacturer...
This is really hard! There always Are unpredictable situations. No Human nor ai can allways manage those. The weather is one interesting aspect. Sometimes there is unpredictable ice on the road. Sometimes wind will carry thick snow to the road. In these cases Human can be better IMHO, but as it has been said. Humans make so Many errors that for example flying is normally safer with autopilot...
 

There actually is a reason people would want to cross there. There's apparently a nice brick walking path crossing the center median of this road at that point, and a bus stop on the other side (with a music theatre behind that). The paved walking path oddly has signs surrounding it telling people not to cross the road there, making the fact that the path exists at all a bit bizarre. And while there is a crosswalk down the road, the road widens out significantly to 7+ uninterrupted lanes at that point, making the 2 or so lanes on either side of the median with the walking path actually appear safer by comparison...

https://t.co/qiBdZ4RIZs


I was going to point this out as well. It's extremely unlikely that the camera data that was released was what the car's algorithms had to work with. That's about the kind of image quality one might expect from a cheap dash cam, and even a typical human driver would have likely been able to see the woman much better than what was shown there. There's even streetlights overhead not too far from where she was crossing. It stands to reason that the company trying to not be held accountable would avoid releasing footage that would show the situation in clearer detail, so one shouldn't take the view from their bargain-bin dash cam as evidence that it would have been difficult for a human to see the woman.

Also, what was the person in the driver's seat doing, who's supposed to be monitoring the situation and taking over in the event of a failure like this? Judging by the video, they were looking down, and not at the road. I'd be interested in finding out whether they were operating their cell-phone at the time. I imagine monitoring a self-driving car could get monotonous, but that's what they are getting paid to be doing.

The fact that they are testing this equipment on SUVs seems a bit questionable too. Why not first get the technology working near-perfectly on lower-profile, lighter-weight vehicles first, and only move to publicly testing it on larger vehicles once the smaller ones are in widespread use, and found to be operating correctly?
 
In Arizona, it's legal to cross mid-block, or where no posted crosswalk exists. Two points go against the woman who was struck however. Pedestrians are required to yield to cars, which clearly she didn't, and it's not legal to walk in posted no-walk zones, which this woman was doing.

The brick pathway is marked as prohibited to pedestrians, so it's it's clearly not a walking path. The signage also instructs to use the crosswalk. The bus station and music hall were on the side of the road the woman was walking from, not her destination. On the other side of the road is desert and a park. Not much to attract somebody unless they are planning to go camping or walk another mile up the road to the next populated area.

One of the purposes of the simple camera footage is to have something for humans to analyze in the event of a systems failure, but it certainly isn't what the car is using to navigate. The car involved in the collision uses a LIDAR system from Velodyne Lidar Inc. to construct a 3D 360° picture, not a simple camera feed. There is also radar radar and optical cameras. Obviously the video that was released is not what the car was using to navigate.

What would you suggest, a motor scooter? Anything big enough to work as a passenger car for the purposes of Uber, and traveling in the range of 40 MPH, is going to kill somebody if it fails to apply brakes and runs them over.

The real question is, were the multiple sensors on the car somehow defeated by the pedestrian or did the car's software fail to take correct action based on good information from the sensors. Either proposition is problematic.
 

I had a look at the area in streetview the other day, and the signage could actually be considered somewhat confusing. Here's one of the signs coming from the side of the median that the woman apparently entered the walkway from...

https://goo.gl/maps/1y7JtoPDXhm

The sign reads "Use Crosswalk" with an arrow pointing in the direction of the diagonal brick footpath that is supposedly not a footpath. Looking at that, it could easily be interpreted that the brick walkway is the "crosswalk" that the sign is referring to, or that the sign might be pointing to a crosswalk on the other end of the walkway. And if she assumed that was the crosswalk that the sign was referring to, she may have even assumed that she also had the right of way, and that cars would be expected to yield to her. Or maybe she was just distracted and didn't see the car coming, and figured any reasonable driver would brake for her. She certainly shares some of the fault in not being observant of her surroundings when crossing the road, but that doesn't dismiss the fact that placing walkways there without proper crosswalks is a terrible design, or that the experimental vehicle should have been properly monitored.

Another thing you'll notice, if you rotate the streetview to the right under the overpass, is that there are pedestrians walking along a sidewalk at the other end, on the median side of the road. In fact, click the view over there, and you'll see that it's a full sidewalk, complete with little alcoves with benches in them, at the exit of a bridge crossing a river...

https://goo.gl/maps/KNLzWXdHaJF2

Does that not look like a place where people are intended to walk? The sidewalk continues under the overpass, and on the other side leads up to that median walkway. You can't even see the sign in question when approaching from that sidewalk, since it's positioned right at the edge of the road facing outward, and you're already on the median by that point, with the sidewalk ending, so anyone there is effectively trapped by no crossing signs that they can't even see.

And if they didn't want people walking there, why on earth did they design it to look like a walkway, and have it connect near the end of a long sidewalk? To make the area look more pedestrian-friendly than it really is? Going by a report published a month ago, out of the 50 US states, Arizona currently has the highest pedestrian fatality rate at nearly double the national average, and a lot of that likely comes down to poor road design. This strange pedestrian walkway that is supposedly not a walkway seems like a good example of that.


I'm referring to potential accidents in general, not just this one. Any small car would likely be safer than an SUV. In this case, the vehicle didn't even seem to apply brakes (nor did the guy who was supposed to be monitoring it), but in general a small car should have less mass and in turn be easier to stop in the event that something goes wrong, and should inflict less damage in an impact. These are still experimental vehicles and are not being used to ferry actual passengers at this time, so there's arguably not much reason they should need to test on public roads with an SUV right now.


One problem with such a comparison is that as far as I know, all of these autonomous vehicles that are being tested on public roads also have a human driver behind the wheel, who is supposed to be actively observing the situation and taking over if something goes wrong, providing additional protection against accidents. In this case, they weren't looking at the road to be able to intervene though.
 
Status
Not open for further replies.