Slashdot Mirror
Human Driver Could Have Avoided Fatal Uber Crash, Experts Say (bloomberg.com)
An anonymous reader shares a report: The pedestrian killed Sunday by a self-driving Uber SUV had crossed at least one open lane of road before being hit, according to a video of the crash that raises new questions about autonomous-vehicle technology. Forensic crash analysts who reviewed the video said a human driver could have responded more quickly to the situation, potentially saving the life of the victim, 49-year-old Elaine Herzberg. Other experts said Uber's self-driving sensors should have detected the pedestrian as she walked a bicycle across the open road at 10 p.m., despite the dark conditions. Herzberg's death is the first major test of a nascent autonomous vehicle industry that has presented the technology as safer than humans who often get distracted while driving. For human driving in the U.S., there's roughly one death every 86 million miles, while autonomous vehicles have driven no more than 15 to 20 million miles in the country so far, according to Morgan Stanley analysts. "As an ever greater number of autonomous vehicles drive ever an ever greater number of miles, investors must contemplate a legal and ethical landscape that may be difficult to predict," the analysts wrote in a research note following the Sunday collision. "The stock market is likely too aggressive on the pace of adoption."
Would a fairer comparison be the number of deaths after humans had only driven 15 to 20 million miles?
Based on the video I saw, she was practically invisible until she entered the car's headlight beams. The road was poorly lit, and she had dark clothing, no reflectors on the bike and no lights.
I don't see how I could have stopped or swerved in time to avoid her in that brief window.
Believe me, I don't care for self-driving cars at all, but I have to remain unbiased here because I know I would have hit her in the same situation.
Be safe out there, people. Put lights on your bike or yourself when you're out there on the road at night.
Maybe a driver could have avoided the accident, but the pedestrian sure as hell could have avoided crossing two lanes of traffic right in front of a speeding car.
Watch the video - she wasn't exactly making herself visible, whereas the car HAD FUCKING HEADLIGHTS and was HIGHLY visible.
Reality found her guilty of stupidity and sentenced her to death. Sentence carried out immediately with no appeal.
Sound cruel? Tough shit. Reality don't care what you feel.
raises new questions about autonomous-vehicle technology.
No, it raises further questions about Uber's poor, perhaps criminally negligent, implementation. In the last year Uber's had more, and more serious, accidents than I think every other driverless program combined. Google/Waymo has been testing in San Francisco - not Tempe - for years with nothing comparable.
.: Semper Absurda
I actually watched the set of videos and there's two major things to note:
First of all the safety or backup driver appeared to be distracted. Although in all fairness if you're suppose to sit there hours on end without taking an active roll at driving this is probably going to happen. This is why google believes in all or nothing approach, half-baked systems are going to get people killed. While this wouldn't save the cyclist from being hurt, quick reflexes may have saved it from being fatal.
Second, LIDAR works by projecting a super high speed panning laser that maps out the 3D spacial environment. It causes the computer to produce a 3D model of the surroundings. This should NOT be affected by the dark! Unless Uber decided not to use LiDAR which would be a dangerous move. If they're using LiDAR the only explanation is the AI image recognition system failed to recognize the cyclist which is weird considering an object that BIG moving should register as a collision threat. Google has noted that in their own self-driving program the computer can sometimes panic over a flying piece of newspaper while a normal driver wouldn't because it looks like an object heavy enough to threaten the car.
....she was practically invisible until she entered the car's headlight beams
Human vision is MUCH more sensitive than cameras. What looks dark in the video wouldn't be so bad to a human. That's why they use all those lights when shooting video.
So, it wasn't as dark as it appears.
Seriously. Pay attention when crossing the road, especially at night.
But the other question would be why didn't she see the oncoming vehicle. It had its lights on and even coming around a bend, the light it threw onto the roadway would be visible long before the car itself appeared.
Even if one party in a collision is not at fault, that doesn't mean they couldn't have avoided it.
politicians are like babies' nappies: they should both be changed regularly and for the same reasons
Kudos to everyone in the last story who commented LiDAR being able to see the pedestrian and the crash being totally avoidable. Comments have also been more accurate than the news in the recent Intel and AMD (non-story) about security.
The fact that the highly moderated comments is more accurate almost any news outlet is why I keep coming back. That and I'm *still* looking for Natalie Portman's brand of hot grits.
-- Political fascism requires a Fuhrer.
Looked like a human pushing a bike.
A human pushing a pram doesn't look like a human either.
Clearly we need more people to die before the statistics can be compared.
video camera's dynamic range is much less than that of a human eye. meaning we can't judge what a human might/could have seen based on this video.
that said, i'm not necessarily in agreement that a human could have avoided a collision is a similar scenario BUT sensors in my view should have noticed the pedestrian. if not that should be considered a fixable flaw.
That's actually true, it is the way statistics work.
Let's say that AI cars have a 'true rate' of 100M miles between fatal accidents, somewhat better than human. Or they could be 50M miles, somewhat worse. The technology is still in development, so who can say.
The fatal accident can happen anywhere in that 100/50 million miles. It could happen mile 1. Mile 20M. Mile 99,999,999.
Spreading that even more, you could have 200M miles with 2 deaths, and have both happen in the first 100M, second 100M, etc... It's really 1/100,000,000th chance of death per mile. You could get multiple deaths, or no deaths.
The only way to nail the real rate down with any degree off accuracy is to have multiple occurrences. Which in this case means more deaths.
I don't read AC A human right
My sister in law just rear ended a semi, an autonomous car would have completely avoided the collision. She was looking to pass and when she checked her mirror the semi slowed down and she ran square into the back of it.
A human driver could have avoided the death, but they also could have made it much worse. These experts are massive wastes of money.
But we did have a lot of people who did not expect to see cars on the road, and instead expected to see a horse. Cars travelled faster than horses, and made different sounds, so if you were not expecting it you could have easily been hit. Take a look at old video of cars and trolleys going down market street in San Francisco from the 1910s and 20s, and you can see how it was a miracle people didn't get mowed down left and right.
Why do you need to kill people to "develop sensors" better? You can model all of this.
Yeah, because getting money makes getting killed by some experiment OK. What an idiot.
I know I sound like an apologist but an "alert" driver could likely avoid most crashes that occur.
Future victims of autonomous car accidents thank you for the abysmally low bar you put on safety levels for autonomous cars. Uber investors thank you.
Laws are rules for the court, but merely a bottom bar to hit for life. Think beyond laws in your actions always.
If there is no need to steer, the attention wanders and it is impossible to stay alert. This was discovered almost 100 years ago in the railroads. The engineer had the exacting task of watching for grades and monitoring speed, especially those days with weak steam locomotives that responded very slowly. Still they would get bored and fall asleep. They invented the dead man's treadle. The engineer must keep it pressed, or the locomotive will stop. Even now there are various techniques to check and keep the engineers alert on railroads.
With that much of history, it is stupid for autonomous cars to just leave the driver there. They should have active devices that do challenge and response to make sure the human operator stays alert. Else it is a waste to put a human being there.
sed -e 's/Chuck Norris/Rajnikant/g' joke > fact
Really? You think hitting ladies with bikes in the middle of an open road is the path to success here?
Laws are rules for the court, but merely a bottom bar to hit for life. Think beyond laws in your actions always.
I've seen the video. I think I would have hit her, but I also think I would have slammed on my breaks before I did. I might have hit her at 20 instead of 40. She might be alive.
Hi! I make Firefox Plug-ins. Check 'em out @ https://addons.mozilla.org/en-US/firefox/addon/youtube-mp3-podcaster/
Good call on saying something that stupid as AC.
If it can be determined that the car *should* have seen her, then what was going on that time that the car didn't see her?
It's a freakin' computer... you can go through its logs and track what it saw and what it didn't see, and figure out based on the logic in the code why it didn't respond to the pedestrian appropriately.
Figure that out, and then add it to repertoire of situations that the car knows about to at least make it safer for the future.
File under 'M' for 'Manic ranting'
So if human driver could have avoided the fatal crash, why not a human pedestrian?
They could have:
1) Worn brighter or reflective clothes while walking at night on a major roadway.
2) Crossed at a crosswalk, where a driver might actually expect and yield to a pedestrian.
3) Illuminated their bicycle with lights and reflectors.
4) Crossed at a location with better lighting.
5) Kept an eye out for oncoming traffic.
Pedestrians have the same brain drivers do and thus the same responsibilities of due diligence.
I don't feel current technology is sufficient to handle self-driving. But this is a clear case where the pedestrian is in the wrong.
Nope, the engineered product needs to be better than humans' performance, e.g. on "sight" and response times. This incident should be immediately recognized as an engineering failure that requires more improvement, not blame shifting to mostly unneeded human. That part is a recipe for disaster, long known in plant operations with far more paid, skilled and trained operators.
Government policies to limit liability (responsibility) are like policies that allow construction of leaky nuclear plants that blowup occasionally due to poor engineering and operations. e.g. disasters in Soviet Union, Japan.
More or less. Continue scoffing at me and shouting me down if it makes you feel better, SDC fanboys, IDGAF.
There are bound to be situations in which a human would react better than an autonomous system. That's not news. The real question is whether there are more per-capita accidents involving human drivers that could have been prevented by an autonomous vehicle or vice-versa. We will likely never get to the point where autonomous vehicles never make a mistake that humans wouldn't. However, when we get to the point where it makes FEWER fatal or potentially fatal mistakes than the average human, that's the cutoff point at which you're net saving lives.
Cars more a lot = more time to get out of the way = far less serious injury when you get hit anyway.
"small sacrifices" like this at least demand full liability. In actuality, normal industrial safety requirements would be all over Uber management for inadequate design, especially since better hardware exists and has existed for decades
obeying the speed limit so useless on the IL Tollwall.
Late nights with low traffic and really good lighting you can fly. No one does the 55 even cops do 75-80 in the 55.
That's not how engineering works. This was an avoidable accident, in large part due to poor instrumentation.
So, something in another lane crosses into yours,
Let's look at it from the computer's perspective.
You can be driving in your lane and have stationary traffic in the next lane (eg. a turning lane). This is not a problem, they are not in your lane.
At the extremes of your sensing range, you see an object in that lane that is not moving towards you. In this case, at that distance a person pushing a bicycle across the lane is - generally - not really approaching you, not if you look at lidar. This is not a problem.
The person enters your lane wearing low-reflectance (to your LIDAR) clothing, pushing an object made of struts and spokes which also doesn't show up very well. Scans now detect an indistinct object in the distance moving into your lane without a defined shape, the bulk of which is only a couple of feet off the ground. Is this water spray or dust, or some other noise? Scans don't seem to indicate that it is particularly solid.
You approach the object at speed and lidar resolves it to be a solid object moving across your lane 20 feet away. This is now very much a problem, but its too late to do anything about it.
You are in a twisty maze of processor lines, all alike.
There is a lot of hype here.
And quietly after the interview Mr Moir though to himself how nice it will be to enjoy his free unlimited Uber account.
Really, the BS he has been coming up with shows that he either thinks of the victim has some kind of human trash who has no worth, or he has a vested interested in protecting Uber - this rhetoric seems unrelated to the facts starting to come out.
not to mention the fact that LIDAR doesnt care about 'shadows' (and there were no shadows, she was on an open straight bit of road - if the Uber was traveling so that it could not in the road it was lit, then it IS at fault).
So long as it hits fewer ladies with bikes than humans do, yes. It doesn't need to be perfect to be ready for adoption. It just needs to be better on average than humans. If 1/10 fatal accidents involving an autonomous vehicle could have been prevented by a human driver, while 2/10 fatal accidents involving human drivers could have been prevented by autonomous vehicles, then the obvious choice is that we should obviously switch to autonomous vehicles for a total reduction in fatal accidents. I absolutely call that progress. Yes, it's going to suck for those 1/10 people who die as a result, but it'll be huge for those 2/10 people who had their lives saved.
All automated driving systems aren't created equal. Waymo and GM's Cruise have redundant braking/power and detection systems.
I would be surprised if Uber even has a fully functioning detection array, let alone any redundancy systems at all.
Human drivers mow down jaywalkers and other pedestrians all the time. Ten others killed just in Phoenix last week alone.
Yes, even though most people overestimate how fast they could react, a human driver who was gripping the wheel tightly and totally focused COULD have avoided the accident. So could a decent AV. But Uber's are some of the worst.
^^^^ THIS ^^^^ The bike rider was an idiot. The car "driver" would not have see her in time I don't think. Also remember the driver was a female. 'nuff said. I though she was a guy until today!
She needs her adrenochrome fix badly.
Doing insanely dangerous things is usually its own punishment.
Let's say that automated cars reach parity with the 'average' human driver. A lot of people consider that acceptable quality so they would be widely deployed. That means a lot of people that would not have been injured by the above average human driver will now be getting mowed down by these 'average' cars. I say to hell with that. These cars need to reach a much higher level of quality. They should be as good as the best human drivers or better before being let loose on the roads.
The human driver wasn't driving, despite the fact that he should have been.
The footage shows him playing on some kind of device constantly for minutes before crash - a phone or a central console or something? His eyes are on the road less than 20% of the time, and his hands aren't on the wheels.
Had the car not been automated, he would have had to be touching the wheel at least, and probably looking at the road rather than his phone.
By no means a guarantee, but having an automated car has thus created two problems - an inattentive human driver who trusts the car implicitly enough to do something that's basically going to be quite obviously fatal, and a car that's not capable of actually performing the duties said guy should be performing.
It's like the worst of both worlds in one. Offloading all the critical functions from the thing that can do it if they bother, to something that can't do it no matter how hard it tries, with no way to pull it back in any reasonable amount of time.
There is so much money to be made that corrupt and incompetent legislatures will allow these machines to kill humans, claiming that the benefit outweighs the loss of human life. This, of course, crosses a line. Machines will be allowed to kill people.
E Proelio Veritas.
Some of the main advantages AVs should have over humans is that they are always alert and that they can detect and react sooner and better to dangers than humans.
There is no good reason why an AV shouldn't have much better night vision than a human relying on headlights. Uber screwed up either on their sensors or on their collision avoidance logic. In the process, they gave the entire industry a black eye. Kudos!
Didn't look like a driver. either. Looked like a Facebook checker.
So in your world, the punishment for jaywalking is the death penalty?
No, the consequence for jaywalking in the middle of the night without wearing any lights or reflective material is winning the Darwin Award for 2018.
I'm not a complete idiot... Some parts are missing.
While there is a sign at that location warning cars to yield to Bikes
https://www.google.com/maps/@3... [Google.com maps street view]
there is also a sign marking the area as a no crossing zone for pedestrians
https://www.google.com/maps/@3... [Google.com maps street view]
From the overhead, we can see that the crosswalk was about 150 feet north of where the woman was hit.
https://www.google.com/maps/@3... [Google.com maps overhead]
The area was not a pedestrian crossing
I'm an experienced individual in the field of control systems.
It's true you want things to fail safe, but it isn't always clear what a fail safe state is.
I'll give you an example.
In a commonly told but apocryphal story, boiler maintenance was being done at the paper mill in Fort Frances Ontario Canada. The design of a boiler (at least a recovery boiler like this) is you have tubes filled with water and steam surrounding a combustion chamber. Usually, you'll have a drum called the mud drum that contains a certain level of water. If that level is too low, that's normally considered an emergency situation. In this case, the maintenance they were doing required the mud drum to be empty, and they were still firing the boiler.
The story goes, a new operator came on shift and saw the mud drum was empty and immediately panicked. The operator immediately opened the water valves wide open (what would normally be considered 'fail safe'), and the boiler immediately exploded.
Why did that happen? What happened is the boiler tubes were red hot and virtually unpressurised. When cold water hit the tubes, the water immediately caused an explosive release of steam which caused an explosion. While the involvement of a person is unusual, boilers routinely experience explosions due to water valves having problems like this. If the boiler was running under normal conditions, perhaps dumping water into the tubes would be a safe option -- cooling everything and getting everything to a zero energy state faster.
So despite the valve opening being what you'd normally consider a 'fail safe' state, in this case it was a dangerous action to take.
Let's assume for a moment that both the car and the driver had perfect vision in that moment, and saw the pedestrian long before the moment of impact.
What is the safest action to take if you see someone crossing the street? Everyone here is immediately saying "obviously slam the brakes and swerve!", but let's think about that for a second. Most people are not going to walk directly into the path of an oncoming vehicle. Even if crossing, you'd expect a person to stop, so you can't necessarily use the fact that there's a person there to predict what's going to happen. By contrast, what happens if you slam the brakes and swerve every time you see someone crossing the street a little too close? If there's a car near you, it could cause an accident. If the person was going to stop, then that person could end up getting hit by your actions where they might not otherwise. The driver or passengers in the car might be injured -- probably for nothing, because 99 times out of 100, the person will stop before the car hits them. Often, the safest act is to do nothing.
Here's where there is a divergence between the powers of an AI, and the powers of a human. An AI sees object 15 travelling at a certain speed at a certain vector. It has to figure out what it can from relatively limited information. By contrast, a human sees a sketchy looking lady walking in a strange way not paying attention. The AI might not recognise there's a threat, whereas the human might recognise something isn't right and take the opportunity to take some of those more aggressive defensive manoeuvres in this case alone.
Our powers of intuition, empathy, and deduction are much more than we give ourselves credit for. We know more than any purpose built AI, and can make connections that no purpose built AI presently can. Humans aren't perfect, but there's reasons why we still have humans involved with even the most high tech processes.
The lane next was empty. A sharp brake and swerve would have missed her.
Deliberately not swerving to miss a pedestrian when otherwise safe should land you in jail.
When looking at the video, at what time did the car start to actually brake? Seems very late to me.
And zero attempt to swerve. I have avoided accidents by swerving, but it does require situational awareness to know when this is safe. One thing a computer should be good at is constantly monitoring what is happening around it.
That video is also dubious. I wonder if the lighting had been deliberately reduced. With ordinary headlights, she would have been more visible sooner. Certainly to human eyes. Or to a decent quality video camera.
So a car that hits ladies with bikes because it can't see them is on track to be an 'improvement'?
Laws are rules for the court, but merely a bottom bar to hit for life. Think beyond laws in your actions always.
It needs to be better than a 95 year old with cataracts!
Laws are rules for the court, but merely a bottom bar to hit for life. Think beyond laws in your actions always.
Auto Cars should be able to see EVERYTHING even without lights. The lights are there only for the squishy people to notice. I want auto cars to succeed, but I fail to see how all those sensors failed to see what auto cars are supposed to be good at, which is noticing objects using hundreds of sensors so that they can prevent accidents. Otherwise, what is the point of auto cars?
From watching the video in this case, this woman didn't "appear suddenly" in front of the car. She appears to have been ambling over from the median, a full lane away. She didn't drop out of the sky or jump from behind a parked truck.
I can't imagine why she didn't react in some obvious way even when the car was almost on top of her - the full toxicology report might be revealing.
Why is there an "insightful" mod and why isn't it "-1"? If I wanted insight, I wouldn't be reading
A human driver had a chance to avoid the collision and instead they were not focused on the road. The backup driver only had to take the steering wheel or put their foot on the break pedal.
The follow up article to this indicates that the driver would have, had she not had her face buried in her cellphone. That "driver" is just as negligent in this case as Uber. She had a responsibility to be ready to take over...that was her sole purpose in being in that seat, and she failed.
Just another day in Paradise
Speaking from experience :
A.
I drive quite a lot when on vacations/week-end, including often on nights, including sometime in fog.
- The human supervisor *should* have turned on the high beams. It seems to me that only the low beams were active, reducing the visibility range. (This might have affected the camera part of the sensors). The super visor is supposed to supervise the self-driving car and thus should be able to see in order to anticipate and compensate bugs, instead of relying the whole thing to work.
- The human supervisor *should* have instructed the car to drive at a speed within the supervisor's visible range (with the low beam only, the visibility is extremely short, the speed should have been kept low).
- The human supervisor *should* have kept eyes on the road (Uber is testing a new technology, bugs are bound to happen.
Definitely the supervisor was doing a couple of things wrong. But even if all the above were followed, that probably wouldn't have saved the bike rider.
B.
I bike to work almost every single working day (welcome to europe), I'm used to bike at night, etc.
- She was wearing dark clothes. It's not a problem per se but you have to keep in mind you're a bit less visible. (Some people here around even where reflective jackets when biking).
- She didn't have any reflector on the bike. That makes the bikes drastically less visible. Usually most bike riders have a good quantity of retroflective reflectors on the bike (plastic on the wheel spokes, sticky bands on the bike body), etc.
- She had absolutely no light. That's a very high danger of collision. Enough for cops here around to pull you if they catch you with an unlit bike. Nearly every one will use a good battery-powered headlight/taillight, often a blinking tail light (legally dubious but every one use them for visibility and police tolerate them). Some bike rider look for "always on" solution to be better visible (magnetic induction tiny lights that are at least visible, even if not very good at lighting. Or on-axis dynamo that powers good lights). Some almost turn their bikes into christmas-tree like light shows just for lulz.
- No helmet (could help diminish the results of an impact, also most of them have reflectors, and some even feature built-in lights).
- She should have seen the headlights from far away. If she counted on the car slowing down, standard practice (massively advertised at the beginning of each school year here around) require to establish eye contact (to make sure that the driver has seen you and will slown down) which is impossible to do by night.
- In the absence of eye contact, she should have waited for obvious signs that the driver will slow down (e.g.: the driver already starting to slow down and blinking the high beams to ack).
- Other wise she should have assumed to not having been seen (specially given the clothes she's wearing).
With my experience with bike I would never attempt to cross the way she did given all the above. That looks absolutely suicidal to me, there's now sane way to expect even a well behaving driver to have avoided the collision.
There must be some reason for her absolutely not paying attention : - being way too much absorbed into some phone conversation over earphones ? - emotionally distressed and not able to focus and pay attention ? - drunk ? (which here around could be a reason for the police to take away your *other* licenses : car driver, boat, etc.)
"Sufficiently advanced satire is indistinguishable from reality." - [Tips: 1DrYakQDKCQ6y52z6QbnkxHXAocMZJE61o ]
this was a death caused by pure negligence on the dead person's part
Negligence is the wrong term, and anyway, the death was caused by multiple factors including the inability of the automated car to detect and avoid the obstruction in the road.
Get a fucking grip, kid
Wait? You're 15?
I think Trump is a buffoon and in the running for worst leader of a free country in modern times, but I think even Trump would have avoided that pedestrian. The Uber-released video is utter shit and makes the situation look MUCH darker than it really is.
Your premise is that a human can effectively take over in the middle of a fast moving emergency situation. This premise is what is false.
The 'average' accident rate of 1 in 86 million miles can be thought of as the probability of an accident happening. One good driver may go 200 million miles without an accident. Another good driver may drive only 10 miles and have an accident. A single accident occurring after 20 or even 500 million miles means (no pun intended) essentially nothing. An average over many accidents of 20 million miles would be cause for alarm. Regardless, Uber does seem to be taking this accident seriously. P.S. If driving, I could not have reacted quickly enough to avoid this tragedy.
Human Driver Could Have Avoided Fatal Uber Crash, Experts Say
"Could have" isn't the same as "would have" - that in hindsight you can imagine a different reaction doesn't mean you could have had that different reaction in this exact situation.
Ken
What, exactly, did the victim in this accident do to avoid getting run over by a car?
She wore dark clothes,
She had no lights or reflectors,
She wasn't looking where she was going,
She walked into the middle of the road.
It's fair to give the victim a slightly greater responsibility for the accident.
Ken
I know I sound like an apologist but an "alert" driver could likely avoid most crashes that occur.
Yes, and that HAS to be the standard for technology.
You do realize that it's kind of silly to bring up the specter of "criminal negligence" when the Tempe police cleared them of wrongdoing, right?
And if the Tempe police say it, it must be true. /s
The video shows a situation in which a human would have at least tried to react. Uber's car didn't even start slowing, and it has LIDAR. Contrary to what you seem to believe, no current self-driving car technology employs computer vision using a visible light camera to guide the car. Instead, they use IR-band LIDAR, detailed mapping data, and other sensor packages common in existing assisted-navigation / crash avoidance systems (e.g. small RADARs).
Anyway, it's confirmed - Uber's program is far less safe, and less advanced, than Waymo's.
.: Semper Absurda