Self Driving Uber Kills Pedestrian

It will be interesting to see how the autonomous vehicles deal with the traffic speed laws. If code is written to purposely break a law, it raises new issues that I don't recall seeing before. Code is often held to a higher standard than people (not that it always succeeds of course).

At one point I heard trains used to have very accurate speed logging systems, every time there was a collision the courts called for it, and if the train was exceeding the speed limit at all it had consequences against them, so they took the logging out. An autonomous vehicle exceeding the speed limit, and recording this information, would seem to put the vehicle more at-fault from the start.

Of course the other issue is knowing what the speed limit is. The GPS speed limit info is often missing or wrong when compared to the signage.

https://www.wired.com/story/uber-self-driving-car-crash-arizona-pedestrian/

Here's a nice example of how people over-react. Too bad this is from a sitting U.S. Senator.

“This tragic incident makes clear that autonomous vehicle technology has a long way to go before it is truly safe for the passengers, pedestrians, and drivers who share America’s roads,” US Senator Richard Blumenthal (D-Connecticut) said in a statement. “In our haste to enable innovation, we cannot forget basic safety.”

wturber said:

US Senator Richard Blumenthal (D-Connecticut) said in a statement. “In our haste to enable innovation, we cannot forget basic safety.”

Click to expand...

He's already forgotten basic public safety if he drives a car among people.

Like when a cyclist-pedestrian collision results in a fatality, this Uber crash is news because it's rare. In the meantime, we sacrifice dozens of lives to normal dipshit-guided cars every single day.

If he wants to take action, he should start with the lowest hanging fruit, which is to start revoking driver's licenses from people who've been at fault in crashes or convicted of reckless driving or speeding.

Everywhere I've driven in my life, you can catch more trouble by observing the speed limit than by violating it-- including increasing your chances of being pulled over by a cop who's supposed to enforce the law.

wturber said:

https://www.nytimes.com/2018/03/19/technology/uber-driverless-fatality.html

Lots more information needed before we can better understand the situation and circumstance.

Click to expand...

Gonna guess it was somebuddy operating a large, heavy vehicle (watt was hard to pedal) that was using an exploding, infernal combustion engine spewing out poisonous "greenhouse" waste gases. At least if it was bettery-electric motor powered, the poor pedestrian would have been hit/run over/killed in a more "Eco-Friendly" way. [Sigh]

Here is the dashcam video of the incident:
[youtube]EPvSS5PhO7M[/youtube]

Wow. Driver appeared to be reading his phone or tablet perhaps. What do Uber's rules require the human monitor to be doing?

One problem here is judging the visibility of the cyclist by a dashcam. I know when viewing GoPro footage from my ebike rides that I could see things at far greater distance than the camera. Things that appeared suddenly in the video had been visible to me for a much longer time. The human eye is pretty good at moderate distances, wide angle lenses projecting onto modest number of pixels not so much.

Very sad.

Two things to consider about the footage itself. The video will exaggerate the darkness because of the limits of the camera sensor and the limits of the highly compressed video file format. It will also exaggerate the impression of distance because the camera lens is fairly wide angle. I suggest looking at the lane stripes to help put the distance into better perspective.

I think a very alert driver might have been able to avoid this or at least hit the brakes if they had decent night vision as well as well aimed headlights without the modern sharp horizon cut-off that this vehicle appears to have. IMO, the woman "appears" suddenly in part due to that sharp headlight horizon cut-off that is typical these days - though the very dark shirt or jacket doesn't help things much. I think it is pretty unlikely that the typical driver would have seen the woman in time. The bike had no side reflectors or lights and the woman was wearing generally dark clothes. The bike and woman didn't provide a very effective reflective target for the Uber car's LiDAR either.

This a case where the lights might have been on high beam had the human been driving. That might have made a big difference.

That is absolutely shit house, the car doesn't even seem to slow down.
There would be a ton of "see in the dark sensors", like infrared and lidar and they totally failed.
This person died because of corporate greed, this is how I fear dying on the road as well.


Seems clear to me if someone wants to commit suicide, looks like a guaranteed convenient way is to merely stand in the road 50meters in front of an Uber car and it will run you over, who needs trains anymore.
Like I said on my previous post on the topic, Uber LOST 4.5billion dollars in 2017 and are willing to do and risk ANYTHING to make the company profitable, their best bet is driverless cars, at the risk of everyone else's safety.
https://endless-sphere.com/forums/viewtopic.php?f=7&t=93360#p1366155

Alan B said:

This a case where the lights might have been on high beam had the human been driving. That might have made a big difference.

Click to expand...

Maybe. But the Uber was approaching an intersection, and there were streetlights - so a reasonable driver would be getting ready to dim them pretty soon. If the woman would have simply looked in the direction of oncoming traffic - street crossing 101 - she'd have stopped and all would have been fine. I see people crossing streets without looking and it continues to amaze me.

It is hard to tell from the video, but under those conditions it doesn't appear that the low beams were adequate for the speed.

I don't recall any legal requirement to dim headlights in the absence of oncoming traffic (or following traffic). Intersections or street lights are not necessarily triggers for low beam lights.

What kind of sensors does the uber self driving vehicle have?

https://www.techradar.com/news/uber-self-driving-cars

Arizona
Headlights
Headlights must be in use from sunset to sunrise.
Headlights should be used in snow and ice.

High Beams
High beams must be dimmed when approaching traffic is within 500 feet.
High beams must also be dimmed when following within 200 feet of another vehicle.
High beams should not be used in fog.
.................................................................................

Here's a Google Map link that puts you pretty darned close to where the accident occurred. This is easily confirmed by the storm drains under the sidewalk and the transition of the bike lane stripe from solid to dashed for the upcoming right turn lane. That should give you a sense of how close the intersection really is. Like with the dash cam, this Google Maps view is very wide angle and makes things look farther away than they are. If you pivot to the left, you will see the median from which the woman was crossing. Now note the red pavement. Tempe has apparently built sidewalks in this median that go nowhere. In fact, they've even posted signs telling pedestrians that they should cross at the crosswalk. Sound advise for sure, but they shouldn't be on these sidewalks to begin with. There is no appropriate and legal way to even get there. So why build such slices of pavement? The city clearly knows that these strips are going to attract pedestrians. That's demonstrated by the presence of the signs. Zoom out to a sky view and you can see these two strips.

Note also that the street lights are not placed where pedestrians would likely illegally cross.

I'm not sure if these short URLS expire, so I'm pasting a short and the full for posterity.
https://goo.gl/nA2kai

https://www.google.com/maps/@33.4364401,-111.9425695,3a,60y,320.48h,71.71t/data=!3m6!1e1!3m4!1snSB0UUiEcTSWEkB0D91vOQ!2e0!7i13312!8i6656

TheBeastie said:

There would be a ton of "see in the dark sensors", like infrared and lidar and they totally failed.
This person died because of corporate greed, this is how I fear dying on the road as well.

Click to expand...

You are far more likely to be hit and killed by a human driver - and that would be true even if the ratio was 50/50 autonomous car vs regular driver.

It's like dying in an airplane crash vs. dying in a car crash. You are far more likely to die in a car crash, but most people are more afraid of the plane crash. Just seems scarier to them.

billvon said:

TheBeastie said:

There would be a ton of "see in the dark sensors", like infrared and lidar and they totally failed.
This person died because of corporate greed, this is how I fear dying on the road as well.

Click to expand...

You are far more likely to be hit and killed by a human driver - and that would be true even if the ratio was 50/50 autonomous car vs regular driver.

Click to expand...

Yeah. I don't buy the "corporate greed" bit. Poor design at this stage could be crippling to their efforts. I'd have to see a much better case made to buy into that theory. Besides, the poor woman who was killed was clearly crossing illegally and not paying attention to traffic. So at worst, there are multiple contributing causes here.

But I'm also not sure that we have enough data on the current state of self-driving cars to know that they are on average better drivers than human beans. I believe that should be the case at some time in the future. I just don't know if that time is now. The more I look at this case and the more I paid attention while driving home this evening, the more I think an attentive driver would have seen this woman crossing and would had time to at least hit the brakes. She had been in the street long enough to have almost finished crossing three traffic lanes. That means she was in the roadway for almost 8 seconds. Lanes here are about 12 feet wide and she had crossed two of them by the time she became visible directly in the car's headlights. She was partially visible in the car's headlights for about 1.5 seconds starting when the headlights lit up her feet. Maybe the typical driver wouldn't have seen her in time to hit the brakes, but I think some drivers would have.

While she was wearing a black shirt or jacket, she wasn't that far away from two nearby streetlights. I don't think she was as invisible to human eyes as the video suggests. I almost drove down tonight to get an idea of how dark that area really is at night. It surely isn't as pitch black as the video shows.

If this article is to be believed, regular Lidar has a hard time seeing a person with dark clothes on. No surprise there really.
https://spectrum.ieee.org/cars-that-think/transportation/sensors/22yearold-lidar-whiz-claims-breakthrough

More current info. https://www.wired.com/story/lidar-self-driving-cars-luminar-video/

wturber said:

So why build such slices of pavement?

Click to expand...

Tempe isnt' the only place that does that--it's common enough around the valley (at least in areas I've ridden over the years, which is most of it except Scottsdale and the "new" cities springing up "outside" the valley).

Also, Phoenix has quite a bit of marked bike lanes that go very short distances and disappear, though not quite the same situation as that paved area.


Regarding visibility: Around Metrocenter area in general (my commute and shopping area) and inside the mall streets in particular, it's very common for pedestrians to just walk out into the street whether traffic is coming or not. Sometimes they look first, sometimes they don't. Some of these streets are 35MPH+ speed limits, and they are generally curved, as Metro Parkway itself is a big oval around the mall. Few of them are wearing anything high-visibility, and quite a lot are in dark clothing. Most of the cyclists don't have reflectors or lights, and a significant number (especially those on BMX bikes) have darkened the entire bike, including rims and spokes, making it much harder to see them at all.


In dusk (or dawn) conditions, there are a number of places where it is VERY hard to see anyone crossing, because of the low-contrast conditions and the headlights of vehicles in the oncoming lane.

I've used quite a lot of different headlights over the years, including various car headlights, and most of what I've used is wide-beam (both vertically and horizontally) to make me as visible as possible from as many angles as possible. Regardless of which lights, brightness or wideness of angle, the pedestrians (or cyclists) are not any different in how visible they are under those conditions.

It does make a difference (sometimes a lot) in complete darkness, where there is a lot of contrast between objects closer to the lights and those farther away.


Over the years (decades, actually) riding bikes pedal and electric, I've had a number of very close calls with pedestrians in those conditions, usually becuase they step off the sidewalk right in front of me--which is why if I have any choice in it I don't ride in the bike lane but rather in the traffic lane, when these conditions exist, and in certain areas I'll slow down to 15MPH (or less) rather than 20MPH just in case as I approach, under those conditions. Sometimes there's other traffic behind me that doesn't consider the potential problem, and in some cases drivers gun it and go around me, sometimes then screeching their brakes and swerving into oncoming traffic to avoid the pedestrian or cyclist that was crossing (as I expected but couldn't see until that moment, which was why I was going slow). So far havent' seen anyone get hit, or any traffic collisions from it, but it's been REALLY close more often than I'd ever like to see.

Regarding how people and computers differ: A human driver can recognize a person or whatever in a place they shouldn't be at that moment by extremely tiny cues (which is what let me not hit any of the ones I had close encounters with). Whether they can do anything about it in the time available, or whether they will even try, well, that's another thing, that varies from incident to incident. But even just a pattern of movement in even the lowest of contrast of lighting can be enough for the human brain to realize there's something non-car in the lane and alert the rest of the brain to do whatever it's going to do about it.

A computer doesn't yet have the same capability, mostly because the vision systems themselves used on AVs generally don't have anywhere near the capability of the human eye. The systems can see stuff we can't, and can see in ways we can't, but they can't see the things that the human brain has evolved to recognize, so we need to develop systems that do this kind of motion-pattern sensing in a different way.


FWIW, of the people I have known in robotics in general, there's two types--teh first tries to emulate exactly how we do things, sensorily and in processing that data. The second tries to invent a completely different way that is task-oriented, but doesn't include any "extraneous" processing or sensory ability, so if anything outside the very specific task comes up, the program has no way of dealing with it--it doesnt' even know it is happening, and just keeps trying to do what it was already doing.

AVs need a third approach, that takes the best of both of those, and some other thing I haven't thought of.

amberwolf said:

Regarding how people and computers differ: A human driver can recognize a person or whatever in a place they shouldn't be at that moment by extremely tiny cues (which is what let me not hit any of the ones I had close encounters with). Whether they can do anything about it in the time available, or whether they will even try, well, that's another thing, that varies from incident to incident. But even just a pattern of movement in even the lowest of contrast of lighting can be enough for the human brain to realize there's something non-car in the lane and alert the rest of the brain to do whatever it's going to do about it.

Click to expand...

Yes. I've spotted pedestrians wearing dark clothes even though I really couldn't "see" them. What I saw was that they eclipsed some distant lights ... a sketchy and partial silhouette, but enough to get my attention. As for patterns of movement, sometimes it is the pattern of movement of other vehicles whose drivers have seen something you haven't - so maybe they aren't pulling out as quickly as you'd expect or are slowing down for no clear reason.

The closest I've come to having a pedestrian step in front of me on my ebike was when a javelina stepped out of the darkness right into the path of my bike. There was no way I could react in time. Fortunately, my light startled the javelina and he/she/it bolted quickly out of my path. That's one reason I got the new "motorcycle" LED light. I casts more light to the sides so I will hopefully pick things like that up earlier.

you can see far ahead an oncoming car's headlights comes out diagonally from the taillight of the car ahead traveling in the same direction. this happens in the first second of the video at the same moment the woman/bike was in the shadows

this was likely "trippy" for the neural net and "consumed" its "attention" (focused on the far field)

tripping out: https://www.theverge.com/2017/4/12/15271874/ai-adversarial-images-fooling-attacks-artificial-intelligence

either way
uber's CEO deserves prison time and uber should be served with massive fines

wturber said:

https://www.nytimes.com/2018/03/19/technology/uber-driverless-fatality.html

Lots more information needed before we can better understand the situation and circumstance.

Click to expand...

Hehe... Trying to cross a busy highway, at night, dragging a bicycle... seems like the usual SHT... Stupid Human Tricks... [sigh]

It's starting already...

“Intelligent future AI will probably develop a drive to survive and acquire more resources as a step toward accomplishing whatever goal it has, because surviving and having more resources will increase its chances of accomplishing that other goal,”

Click to expand...

Stephen Hawking

Seen in ES Newz...:

https://endless-sphere.com/forums/viewtopic.php?f=3&t=57933&p=1366722#p1366722

:wink:

wturber said:

Yeah. I don't buy the "corporate greed" bit. Poor design at this stage could be crippling to their efforts. I'd have to see a much better case made to buy into that theory. Besides, the poor woman who was killed was clearly crossing illegally and not paying attention to traffic. So at worst, there are multiple contributing causes here.

Click to expand...

Agreed there. Given the amount of money pouring into basic research into autonomous driving systems, it's hard to argue that greed is a motivator.

She was partially visible in the car's headlights for about 1.5 seconds starting when the headlights lit up her feet. Maybe the typical driver wouldn't have seen her in time to hit the brakes, but I think some drivers would have.

Click to expand...

Also agreed. And further I would argue that most autonomous vehicles would have as well. There's a huge amount that goes into the ability to detect an unusual obstacle and process it. The LIDAR scan pattern may have just missed a detection on scan 2031, then on scan 2032 it was too late to stop - so a different timing might have made it detectable. Or the scan may have detected something, only to have the matching algorithm misclassify it as blowing trash or something like that. Studying the logs will allow engineers to figure out what happened and make the changes needed to reduce the risk of that in the future.

billvon said:

Also agreed. And further I would argue that most autonomous vehicles would have as well. There's a huge amount that goes into the ability to detect an unusual obstacle and process it. The LIDAR scan pattern may have just missed a detection on scan 2031, then on scan 2032 it was too late to stop - so a different timing might have made it detectable. Or the scan may have detected something, only to have the matching algorithm misclassify it as blowing trash or something like that. Studying the logs will allow engineers to figure out what happened and make the changes needed to reduce the risk of that in the future.

Click to expand...

The more I think about this video the more I'm beginning to think there was a system failure. The time between scans in all the demos I've seen is milliseconds. I see the woman taking multiple steps in this video. Also, it seems like the system should have detected her at least once she was within 100 feet. That should be time enough for the vehicle to at least apply the brakes, if not stop in time. They keep saying that the vehicle was going about 40 mph, but I seem to recall a mention of 38 mph in a 35 mph zone. The average braking distance for a car at 40 mph is 80 feet. So you'd expect a Volvo to stop even shorter than that given Volvo's claims of building safe cars and the car not quite going 40 mph. But the car never appears to apply the brakes. I'm pretty sure these cars were designed specifically to deal with situations like this precisely because we know that people will do careless things like not cross at crosswalks and not look for oncoming traffic. So at this point it looks like a system failure of some kind to me.

Just did some checking and 38 mph = about 55.7 feet per second. That means the car was about 80 feet away when the woman's feet became visible. So even a moderately alert human bean who only saw her when her feet showed up in the headlights would probably we on the brakes after about a second of elapsed time and would have at least been on the brakes scrubbing off speed before hitting the unfortunate woman.

Would agree that an attentive* human driver would probably have seen her. Shadows just don't work like it appears on the camera footage, the only way you'd see nothing at all would be if oncoming headlights were dazzling you. The self-driving sensors can give superior vision in some circumstances - humans just can't see through fog, for example. But, the human brain is very well evolved to spot the slightest possible movement of a object in dark places and we err towards false alarms rather than ignoring things.


*Operative word and often not representative of the majority of drivers

Do they have a driving test for autonomous vehicle control systems? Seems like there should be standardized testing under development by the government or some independent testing lab franchised by the government that autonomous vehicles would have to pass before being allowed on the road.