Slashdot Mirror
Police Release First Video From Inside the Uber Self-Driving Car That Killed a Pedestrian (recode.net)
An anonymous reader quotes a report from Recode: Three days after an Uber self-driving vehicle fatally crashed into a pedestrian in Tempe, Ariz., police have released video footage of what the vehicle saw with its cameras moments before running the woman over, and what happened inside the vehicle, where an operator was at the wheel. The video footage does not conclusively show who is at fault. However, it seems to confirm initial reports from the Tempe police that Herzberg appeared suddenly. It also showed the vehicle operator behind the wheel intermittently looking down while the car was driving itself.
THE BUGGER
I don't understand people like this. It's dark, you aren't lit, you're crossing a road with a large, fast moving, well lit hard to miss cars. And you can't be bothered to look for oncoming traffic. Only saw the video twice on the news but it looks like she never knew the car was there.
That said, I'm glad I was correct in my knowledge of what those "safety drivers" actually do all day.
This is a good example of why visual sensors are insufficient for autonomous driving.
Yes, this was hard to see using passive techniques with visible light (ie, your eyes), but WTF, the person wasn't sprinting or jumping off the curb, something active like LIDAR should have had no troubles spotting this.
https://www.random-science-tools.com/physics/stopping-distance.htm
see if you'd be able to stop, I know I wouldn't have, maybe swerve and that's a big maybe.
The pedestrian stepped in front of the moving car.
Sometimes people do things for which technology ( or the user of that technology ) has no way to compensate.
Though some attorney may well milk settlement money out of this, it really does seem the pedestrian is 100% at fault.
It's time pedestrians started taking responsibility for their part of the safety equation. It's not realistic to expect cars to always compensate for poor choices made by pedestrians. Of course anyone who walks in New York City will already know this is true, because if they didn't know it they'd probably already have been hit by a vehicle.
I think we will make progress on these issues when we collectively stop pretending that "operator inattention" is the intended result of using of automated cars, not an unwanted by-product.
Hard to blame technology in this case. I can't imagine not hitting someone in that circumstance.
I've seen this happen on huge college campuses as well. Legions of kids crossing streets while paying zero attention to the potential for oncoming traffic. Usually it's because their face is buried in their phone, but sometimes it's not, and they literally step right off the curb into traffic for seemingly no reason. It might make me sound like an old guy but my generation had a healthy fear of death by car instilled into it (by our parents and guardians) which seems to be sadly lacking these days. It's amazing that more people aren't routinely run down.
What the fuck is this? Is it a joke?
Where is a video with overlay of objects detected by car sensors?
I stand by my post in the last topic :
https://slashdot.org/comments....
(You people are CRAZY with your laws and cars, pedestrians should NOT have right of way at all times as it seems you do, or at least most of you behaved while I was there)
That being said, someone speculated this was right near a late night club in a boring area, only fun thing to do is drink there? So maybe drunk.
I'll tell you a few things, they crossed in the WORST spot, just IN the dark part after some light, holy crap does she come out of nowhere.
Also, texting or not, there's NO CHANCE I personally wouldn't have hit that person, based on that video. That person crossing is an idiot. The driver is going to get slammed for this but they couldn't have averted it.
(that being said, video ! = eyes, and generally you can see better than a camera can)
Finally, as someone else said, radar, lidar, laser, what about all the other fancy tech these things should have? (Does it?) because based on those technologies, the speed the person is crossing the road, the direction they're going, the clearness of the road in front of them? Surely some of these features should've picked up the idiot cyclist?
Bonus finally: It's kind of on the cyclist here to also be the one looking for headlights. The car is lit up and presumably noisy, on a motorway, the car can't see you, you don't have bloody headlights, that person should've been able to look up the road, see headlights, not cross?
Clear sky, dry road, no other cars, straight well-marked pavement...
The road conditions could not get more ideal for avoiding this accident. It's pretty incredible that the autonomous car would fail so spectacularly.
p.s. that camera footage is suspiciously bad, even a $50 dashcam would do better
I had heard reports that the video showed her popping out of no where. Absolutely that is not what it shows. It shows here suddenly coming into the headlights lit region, not appearing from behind a bush.
What's the cardinal rule of driving at night or snow storms? NEVER outdrive the range of your headlights. That is, your stopping distance abolutely positively has to be within your range of sight. Anything else is completely irresponsible.
So that's clearly what happened here. The woman in the video just appears like magic in a couple of frames from dark to in the head light. That means the leading edge of the headlight zone was something less than 1/2 of a second from impact. No way can you stop in that time.
This is defacto outdriving your headlights. Uber is guilty. Case closed.
Now moving on to technical details this also shows. I think part of this is that the dymanic range of the camera sucks. I am fairly sure my own eyes would have been able to see further into the dark. Those black pixels ar not just dark they are completely saturated on the dark end. Nothing is resolvable in them which is why the appearance time is so short. This is a serious problem for all systems as the dynamic range of most cameras is very limited, especially when were dealing with 1/R^4 light fall off ( 1/^R^2 light outbound and then 1/R^2 reflected. Thus a 256 bit sensor is effectively a 16 bit dynamic range sensor. And if you were to account for glints and such then it's even less. No wonder she pops out.
Secondly, where the hell was the lidar here? Shouldn't that have spotter her?
Uber is flagrantly at fault.
Some drink at the fountain of knowledge. Others just gargle.
I am sure there are many instances that self-driving cars have braked and saved lives that otherwise would have been an injury or death. Such incidents never make to the front pages.
Exactly what I expected to see....
Someone walking a bike.
At night.
No streetlights.
No backlighting at all.
Wearing black top and dark pants.
With no lights at all on the bike.
No lights on the person.
Not in a crosswalk.
Apparently not looking.
About 2 seconds of visibility.
The pedestrian is almost 100% wrong in every possible way. I don't see how this could be ANY human driver's fault, had a human been driving. As for autonomous, I guess it depends on what sensors. Could their system have had an infrared camera or other sensor that could have seen the wreckless pedestrian sooner than was evident in [human] visible light? That would have been nice. But does that make the pedestrian less at fault? I think not.
Human eyes can see further in the dark. After all the woman herself was in the dark and surely she was able to see where she was going. Were good at total darkness. because it's not total and we don't have the near field of the headlight reflections blinding us to the dark regions like the uber had.
It was obviously the pedestrian fault. But:
The recording shows, that there are at least 2 seconds since first visible parts of the person were on screen to the hit. The crash was possibly avoidable if the driver was human (depending on the driver's speed of reaction, tiredness etc). The car did not try to brake nor change its path until the very end. As a driver I would have 2 seconds to react. Lidar should have seen the person even before the light hit her shoes.
... however the test driver did not really pay attention.
Being test driver is obviously a fucked up job. 99% is killing time and 1% is killing time.
In Germany there is not one test driver but 3 ... one who would react if something goes wrong and 2 to write protocols about notable stuff.
In this case it is notable that the lights are configured incorrect. They barely shine 15 yards ahead, that is definitely wrong, and a driver or the automatic driving system should adjust speed to about 1/3rd of what it was driving.
Cost free eBook I read (by iBook/Kobo/Amazon/ObookO/Gutenberg etc.): "The Green Odyssey" by Philip Jose Farmer.
I'm sure serveral automation fanbois got a hard reality check today. Tough love.
Laws are rules for the court, but merely a bottom bar to hit for life. Think beyond laws in your actions always.
Yeah, he’s there baby sitting the new tech but, assuming the vehicle didn’t slow down, could he have intervened if he wasn’t distracted? He appears to be reading something since he smiles after looking at whatever he’s holding below the camera view. My money is on that guy getting an NTSB finger pointed at him.
Chewbacon
The Bible is like Wikipedia: written by a bunch of people and verifiable by questionable sources.
True she comes out of nowhere on the video, but that's a really crappy video. She was walking slowing and already in the car's lane when the headlights hit her, even if she had been stationary the result would have been the same.
Of course a human driver could have hit her as well, but I suspect that most often a human driver would have seen her far enough ahead to stop or at least swerve enough to avoid her (of course most Ubers might have as well).
I'm curious if that's the only video available since decent cameras are not that expensive, and I'd expect the car to have several cameras at different contrast levels.
I stole this Sig
What a horrible thing to watch. That woman would be alive had there been a decent human driver at the controls. Having been in similar situations with dogs and wildlife several times, her death was avoidable. Sudden hard braking and a slight swerve to the left would have almost certainly allowed her to survive the encounter, even if the vehicle had struck the bicycle's rear. Bicycle wheels, being mostly spokes, are fragile and give way easily. Too bad the human driver didn't notice until it was much too late (previous reports stated that the vehicle never braked at all). We should also consider that the video is likely taken from a vantage point different from the operator's and the operator may well have had a much better view. The video shown depicts an unnatural range for the headlights, compared to all the vehicles I have driven (many dozens), which cast light much further and higher than what the video depicts. I hope they get serious about the safety drivers that they put on board these test drives, too. Also, before you point out that she made a serious mistake, we've all made at least one on our lives; do we all deserve to die as a result? Modern society is about mitigating risk of death when feasible. Policy wise, this might serve as a warning for all of us, lest future law revolve around convenience for the machines. See "The forgotten history of how automakers invented the crime of "jaywalking"".
When the Uber engineer had the choice but was forced to make the made one, it's called murder.
I maintain that a 2 mile stretch of road with a limit of 35 (previously 45mph) thatâ(TM)s eight lanes across should have more than one crosswalk, and probably shouldnâ(TM)t exist in this form at all. I hate Uber and the empire theyâ(TM)ve built on the backs of the working poor, but city planning has to modernise with our tech. The real wonder here is that people arenâ(TM)t killed on that road CONSTANTLY.
Comment removed based on user account deletion
The video seems a little high contrast which might conceal what otherwise might have been detected by a human, also I can tell you my high beam are on anytime I don't have clear depth and peripheral view. I sincerely believe that I would not have hit that walker.
OK. I see all the posts that blame the pedestrian. Shameless mess those are.
I actually drive a car. No really, i actually drive a car. A VERY uncomfortable situation in a car is where the car is speeding into darkness at a rate where I can't react in time if something is in the road. It is a situation I avoid like the plague. I avoid it out of a sense of responsibility. I also avoid it out of a sense of self preservation.
Clearly in the video the car is speeding into darkness at a rate that it can't react fast enough to what becomes visible.
I actually drive a car. DO YOU? If you actually drive a car and you notice you are traveling at a rate where the you can't stop the car in time if there is something in the road do you turn on the brights or slow down? That's what I do. I suspect that for the vast majority of people that's what you do as well.
That's not what the Uber car did. That's not what the uber driver did either. It actually looks like she is fighting sleepiness.
NO WAY IN HELL this accident should be considered acceptable.
Look at it this way. Suppose a large rock rolls from an embankment into the road. Your amazing tech self drive car is traveling at a rate where it can't stop in time if it sees the rock. Your amazing tech self drive car slams into the rock driving the engine rearward against the firewall. The firewall jams against your legs. The fuel line is compromised and the car catches fire. The fire travels rearward and you can't get out as you are pinned by the crumpled firewall.
ARE YOU GOING TO BLAME THE ROCK?
This was an actual event that happened to a couple as told to me by an officer that came upon the scene. They couldn't get the couple out of the car in time. This crap actually happens.
Deer in the road kill people. Trees in the road kill people People in the road kill people.
A car entering darkness at a faster rate than a response can be effected is a bad idea and is always the vehicle operators fault. Always.
Uber has trannies sitting in the drivers seat doing their hormone therapy
Suppose that this was not a self-driving car. You see a video of a driver spending 50% of their time looking down at a (phone, book, video game, etc.) and 50% looking ahead. They look ahead, and suddenly get an OH SH*T look and plow someone down. What would the law say?
1) The pedestrian was negligent.
2) The driver was negligent.
This is contributory negligence, and I don't think the driver would get off with no penalty just because the pedestrian was negligent. This cannot be allowed to continue.
So back to the self-driving part: either the driver thought "Oh, it's a self driving car, I'll play a video game" or Uber said "Monitor this status console here on your lap and just look up every now and then to make sure that you don't plow over someone." The police need to figure that out. If it is the former, the law should do whatever they normally do in cases of contributory negligence. But if it is the latter, then Uber needs to lose their license for testing these cars, and face a big fine.
Dumbass. You should never look into a headlight, you look away from it if you want to see anything other than pitch black on the ground. she's doing what everyone would do short of freezing like a deer looking into your headlights.
That's at least enough time for a real human to begin applying the brakes.
Slowing down by just 5mph would have given the woman a 30% higher chance of surviving.
With real eyes looking and not a camera, you'd be able to see more detail in the shadows. There are street lights there and a human eye has a much greater dynamic range than a camera.
The person behind the wheel looked up and to their left, showing he saw the woman in the other lane before the impact. The camera couldn't see the woman until she was directly in front of the driver's side of the lane, proving a person could have seen her in the shadow where the camera, due to its limited dynamic range, couldn't.
Perhaps Uber should have forked out for HDR cameras.
It is impossible that pedestrian was invisible to the lidar. Emergency braking was disabled because of false positives when detecting objects smaller than cars. The same thing happened in the Phantom ai accident (see crunchbase article). The driver was probably not aware his âoeintelligentâ car was blindly following the map, but he is also at fault.
Was it really this dark? Or did someone darken the video?
The NTSB report on this accident will be interesting, and likely will say things about Uber's design choices NTSB UPDATE: Uber Crash Investigation.
Industrial safety systems are supposed to fail to a safe state. There is often an operator, but his job is recovery after a safety shutdown. The same general approach should apply to cars and safety drivers.
Sounds like she was homeless. She was also pushing 50. I'm guessing she was just out of it. Fatigue will do that to you. By all accounts no drugs were involved.
I can't fault her for the dark cloths. She was homeless. It's not like she had a lot of options. Maybe the solution is a program to give homeless people reflective clothing. Homes too might help. And medical care. Like I said, she might've just been tired. She might also not have been 100% right in the head.
Hi! I make Firefox Plug-ins. Check 'em out @ https://addons.mozilla.org/en-US/firefox/addon/youtube-mp3-podcaster/
There should be a minimum of 2 and they should be given tasks that prevent them from distracting each other. No talking allowed. I cannot imagine any human short of maybe an astronaut with the capacity to remain focused on such a mind numbing job. Heck even when I drive my own car I realize there's huge sections of time where I can't even recall the drive. The only safety there is that because I'm actually controlling the car there some part of my brain reading the road even if it's not my frontal lobe.
Go to a well run public swimming pool and watch the life guards. If they are well trained every couple minutes they will stand up and sequentially point to a dozen different places in the pool. They are using biomechanics to force their attention not to glaze over. You have to do stuff like that and you have to rotate the lifeguards. One safety person in the car at night is nuts.
Some drink at the fountain of knowledge. Others just gargle.
First why is there light shining on the driver? wouldn't that blind them? I don't think that's ambient light because the passenger seat and walls are not illuminated like the driver is.
Second, what is with the coming out of the back of the driver's head? Seriously look at the video.
Some drink at the fountain of knowledge. Others just gargle.
If you framegrab the images and then histogram the light curve it's hard edged at zero. Someone deliberately made the blacks blacker so it seems like no one could have seen her. Perhaps this is an artifact of the video compression algorithm or the camera itself.
Some drink at the fountain of knowledge. Others just gargle.
The car was going way to fast for the distance ahead of it that the headlights were illuminating, They were definitely not providing 4 seconds of visibility ahead of the car
A human being might have got their foot off the accelerator in the time given them - a really great driver might have got his foot to the brake pedal before impact. This is a nightmare scenario for any driver. Oh, and there are vehicles ahead, so it would be illegal to be running high beams. But I agree - detecting things like this is the job of the LIDAR and/or RADAR systems. Why it didn't is really interesting to me - yes, that's the serious problem here.
Prediction for end of Universe #42: Fencepost error in Quantum_bogosort.cpp
It was manslaughter. Lidar should have been able to see her. I think a meat slab MAY have missed her, but maybe not. Deer are common in my neighborhood and I have yet to hit one. A few panic stops, but no contact. And deer basically do exactly what this woman did only faster.
She is coming in from the drivers side and the impact point is on the passenger side. The car was not even trying to stop. It takes about a second walking to get from her point of visibility on the camera (crap quality so a person would have noticed her sooner) to the point of impact. In a second the car could have slowed to maybe 15mph, which might have been enough to save her. And given a person would have likely noticed the reflectors moving on the wheels of the bike, a person would likely have avoided the accident completely. But what is truly obscene is the Lidar system missed her. I checked, Uber supposedly uses a 360 degree lidar system mounted on the roof. And it missed her. Uber's also have a radar system in front, but someone on foot may not be a large enough reflection to trigger a stop. But how did the system not even begin to brake. It mowed her over. The programmer who designed this hot mess needs to go to jail.
The truly sad part is this woman was homeless and has no one advocating for her. This story will go away faster than a school shooting and nothing will change.
It seems this car (the software that is) perhaps was doing the human equivalent of "overdriving its headlights" - i.e., the sensors it had available to it were unable to see far enough ahead to give the software time to react to an obstacle in the road and bring the car to a controlled safe stop. Alternatively, it saw the woman and the bike but failed to classify them correctly for some reason. Either one is a priority 1 bug in my view.
The driver, had he been paying any attention, might also have been guilty of allowing the car to overdrive both its sensors and his own vision and ability. In that case, he should have either taken complete command of the car earlier or manually adjusted the speed (if the UI allows that) while leaving the rest of the driving to the software. Of course, since he wasn't paying attention, about all we can know is that he was guilty of reckless distracted driving.
Yes, the woman was at fault also, but this is exactly the sort of thing that I would expect a self-driving car NOT to do even in a case like this. IFF self driving cars are reliable at these simple things (don't hit stuff in front of you, don't hit buses to your left, and even don't drive under semitrailers that are turning in front of you) am I going to be tolerant of some of the unnecessary delays they will cause in some traffic conditions or odd, rare and complex corner cases where they make an incorrect judgement resulting in an accident.
It seems that this particular case is an obvious test case that one would run long before putting the first car on the public roads so this is likely also a serious QA flaw. It smells a bit of a situation where everyone in the development organization is incented to ship now, debug later. Perhaps companies who are developing self driving cars should hire some engineers "outside the camp" whose sole job is to develop and help execute tests without pressure to "ship now, debug later". Perhaps this independent team should be given about four hours a quarter of every executive, manager, architect, and senior engineer involved in the project (all the way up to and including the CEO) to use as test dummies (after a detailed description of the test, they can opt out -- but that fact will be recorded and will be considered a test failure, it's up to the CEO to then take sole responsibility for declaring that it's okay to ship with that test failure -- such as if the test involved requiring the "test dummy" to jump off the top of a parked truck just into the path of the car when the car was just three feet away and going 60 MPH).
Why is there an "insightful" mod and why isn't it "-1"? If I wanted insight, I wouldn't be reading
The person was actually crossing only a few meters from a streetlight. Roll the video back a few seconds from the impact and you see it on the top right. So it was very likely a well lighted area. The camera doesn't have enough dynamic range to see it, but that the issue of the camera: it seems blinded by the car headlights. It seems very unlikely a human driver wouldn't have seen this person crossing well in time to make a stop.
Yes, I am too stupid to grasp simple physics. Or even complex physics, since that's what I have a master's degree in.
Fixed that for you. A masters degree in physics is the consolation prize they give morons that aren’t smart enough to pass their PhD qualifiers.
Since were talking a self driving driving, reaction time for a computer is far faster than 500ms it’s ns time scales. Making it effectively instantaneous. A car will absolutely will stop in ~10m going from 50 kph from the moment the brakes are applied.
Signed,
Person with an actual PhD in physics.
The hair the hair!!!! What the hell is that?
The technology in the car worked. It didn't emergency brake, swerve, flip and kill the 'driver' while trying to avoid a jaywalking homeless drug addict.
In GOD we trust, all others we monitor.
The video appears to show that Uber's autonomous car failed completely and killed a pedestrian. It shows that the onboard LIDAR system failed to see the woman who had already crossed one lane of traffic and was walking directly into the path of the vehicle. Note that LIDAR does NOT require light/illumination to "see" objects/people. This appears to be a horrible design failure and (to me) it shows that Uber is at fault. (Not to mention the safety driver, who appears to be looking down at a phone.)
Autonomous cars are designed to use LIDAR data to detect and track the movements and trajectories of all objects around the vehicle, in order to avoid hitting anything. In this case, Uber's system failed to do this correctly.
It wouldn't have helped the pedestrian see the car coming given that it was painted in a dark colour (grey), and was coming out from under a bridge.
I think all self-driving cars under test should be white or brightly-coloured. In fact I'd like to see accident statistics for black vs. white cars.
Copy a frame into an image editor and you'll see that the region that's directly in the beam of the headlights is at least a full stop darker than any auto-exposure algorithm would choose. And somehow the regions that should be illuminated by spill light (not to mention the many street lights) are clipped to pure black.
I call bullshit. Go ahead and pull up any nighttime dashcam footage on youtube. Even with a shitty old dashcam you can see the road more than 20 feet ahead. This camera is either running nonstandard settings that do not even attempt to represent what the scene looks like or the video has been edited.
All the modern high-end automotive vision sensors have good low light performance and really cool live HDR capabilities. I bet the pedestrian shows up clear as day in *that* video.
There is a car (I see something resembling headlights) far behind the woman , with roughly the same angular velocity as projected by the woman. It is possible that the machine vision is trained to not take it into account if the data (or a part of it) is coming from a distance.
Bingo Dictionary - Pragmatist, n. A myopic idealist.
Kids crawl before they learn to walk and run. How about we hone this technology for day time use first and solve the darkness issue with more mature technology later? May be make an exception for freeways, and allow it even in dark. But non-freeway streets, just disable it.
Its human nature to not pay attention or have wrong ideas and expectations about right of way, but getting killed for it is very unfair. Humans can make mistakes, machines can't! We have to get to that level of reliability for this self-driving idea to be viable.
That is also one of the reason we dropped first city speed limit to 50 then to 30 kmh - because in city the braking distance due to lack of visibility is very important. I can't judge for the number on really braking but I can vouch that the reaction time is usually above the 750 millisecond range and at 30 kmh this is nearly 5 meter. Or for the imperial unit for the car case we are speaking of : 35 mph for an attentive human reaction time of 750 ms this is 50 feet per second or roughly 42 feet , or 14 yard. So the women would have had no chance to apply brake. The car itself will also take quite some time so even with lidar :
https://www.google.de/url?sa=t...
so even discarding the human and assuming lidar instant response, the car would still need 136 feet to brake, so into yard about 45 yard. Impossible to avoid the woman even if car start braking instantly. That said I would like to know if it did start to brake on its own or not...
C. Sagan : A demon haunted world:
http://www.amazon.com/gp/product/0345409469/
visit randi.org
When humans learn to drive they are taught to use their best judgement in certain situations. Humans also have reflexes, intuition and experience. Machines have none of the above. Machines know the laws and rules, that's what they follow.
People are just acting outraged because they're either are against self driving cars because it takes away their freedom to be complete morons behind the wheel or because it's Uber, which is such a detestable company which pays their drivers badly. Wonder what the public opinon would be if it had been Google or Tesla or any other company experimenting with SDC.
If your vehicle is moving such that your reaction+braking distance is greater than your visibility (whether that's a sensor or an eye), then you're at fault. It's really that simple. You are putting the car into a position where you cannot possibly react to any feasible hazard in the road ahead.
I'm prepared to believe that the video contrast is not reflective of what it looks like from the car, but I've ALWAYS found that dashcams are better than you think and see more than you would, especially in post-analysis.
But the woman isn't DIVING across the road at 100mph. She isn't stepping out from behind parked cars where there's no possibility of being seen. She's walked across the road. The light-beams of the car pick her up. No corrective action is taken until she's already inevitably dead. The car didn't see her. Like the driver didn't see her.
If it was a human driving, that's because "it was dark". Then slow down, so your headlights illuminate the area you're going to need to brake inside.
If it was the computer driving, the sensors weren't able to pick up objects outside a certain range. Then it shouldn't ever be going fast enough to not have a braking distance inside that range.
It's really quite simple.
There's a whole lane to her left. She's at walking speed. You didn't see her despite being "wider" than a person because of the bike. There's a dark spot and your headlights aren't illuminating it and the street lighting isn't great at that point. But no corrective action occurs. The front doesn't dip. The car doesn't slow. The steering doesn't avoid. The speed isn't dropped when visibility drops.
That's an emergency braking scenario, sure. Unexpected pedestrian maybe (but in the UK, you get that kind of thing all the time, and we don't have jaywalking etc. laws). But the fact that even the low-light kit of a CCTV camera can't pick her up until even emergency braking wouldn't suffice means that it was going too fast for the conditions. And that NOTHING appears to change in terms of the car's motion... that indicates absolute failure.
Sorry, no matter how perfect a driver we might think we all are... this is bad driving. That the human "driver" isn't looking is merely incidental in this case, because he couldn't have done anything from that speed - and thus it shouldn't have been AT that speed.
This would be excuseable on a non-lit motorway (designated absolutely no pedestrians, no side-walks, no access to pedestrians, CCTV surveillance and automated warning of "pedestrians in road" for miles if it does happen). But on a lit road, what we'd class as a "dual-carriageway" in the UK, though it might be legal to do quite high speeds, it's always at the driver's discretion and responsibility. And that change from "I can see all the road" to "I can see what's in my beams" should have prompted a slow-down.
And, at the very least, evidence of braking and movement to the left to try to avoid the pedestrian. That camera literally doesn't dip right up to the point it hits her. That means no braking or no suspension. It literally didn't "see" her at all. Not even at the last moment. Not at all until contact occurred. It didn't detect her, even "too late" or try to avoid it. It didn't even know it was going to collide until it actually did.
I hope like hell that whatever transport safety board is responsible demands data from the car about when it detected her and what its expected detection range was at the speed it was doing, and what action was taken before collision.
I also find it suspicious that it cuts out at point of impact (sure, censor it)... that might suggest that maybe the car just kept going and the guy had to stop it.
After reading many comments, here you can test how fast you can stop a car:
http://extrat.liikenneturva.fi/pysahtymismatka-auto/fi/
Sorry, text is in Finnish, but pictures are pretty self explanatory. Speeds are in km/h and distances are in meters.
Reaction time is 0.5s if one is concentrating, 2s during dark. Older people are slower, especially after 50 years. For detecting breaking lights in acar front of you the reaction time is 0.9-1.8 s. If car stopped on the side of the road opens a door, it is detected in 3-4 seconds. If car slows suddenly but breaking lights are broken, reaction time is 6 seconds (reaction times from Finnish government traffic teaching material based on extensive real tests).
https://www.youtube.com/watch?v=HNcqpBnSwps
Did some quick research and Waymo (Google) had done 5 million miles by January this year. Uber had completed less than 2 million miles.
An aside: found an article from last year that the Uber “self-driving” cars were needing intervention approximately every mile:
https://www.recode.net/2017/3/16/14938116/uber-travis-kalanick-self-driving-internal-metrics-slow-progress
To guess there would be around 8 million miles between them by now.
So there has been 1 pedestrian death from 8 million miles driven by “self-driving” cars (the Tesla death doesn't count because it was the “driver” that died and Tesla don't have a self-driving car, they just have radar-assisted cruise control that they dishonesty call “autopilot”).
From a Wired article soon after the Uber kill:
https://www.wired.com/story/uber-self-driving-car-crash-arizona-pedestrian/
“Humans don’t have a great record: Nearly 40,000 people died on American roads last year. Almost 6,000 of them were pedestrians—that’s more than 16 per day.”
I found this from the US Road and traffic authority (with expired SSL cert on site) from 2001, so miles would be more per year now, but it will do for an approximation:
https://www.rita.dot.gov/bts/sites/rita.dot.gov.bts/files/publications/highlights_of_the_2001_national_household_travel_survey/html/section_02.html
"Annually, the total number of vehicle miles traveled in 2001 was nearly 2.3 trillion."
That is about 6.3 billion vehicle miles per day.
So with human drivers it is 16 pedestrians deaths per 6.3 billion miles - one every 393.8 million miles.
The "automated" car death rate so far (approx. 1 pedestrian death every 8 million miles) is 65.6 times worse.
So if we were to magically replace *all* the human driven cars with current technology “self-driving” cars (personally I think the transition period would be a very dangerous time) the US daily pedestrian death toll would go from 16 pedestrian deaths a day to around:
1,050 pedestrian deaths per day, over 360,000 per year.
No reasonable person would find this remotely acceptable. This makes the human “record” look positively amazing.
This is assuming there are no deaths from car verses car accidents (they should be low with the “self-driving” cars using Vehicle 2 Vehicle technology). Whether that is true is yet to be seen.
Autonomous cars must have similar or better safety record as AVERAGE human drivers. That is rational.
When they are tired or a bit sick.
Then they can quickly be slow and sloppy.
That is why we see lots of disastrous accidents every year. Entire families being wiped out by crashing on the rear of a traffic jam completely unbraked. At daylight, because the driver was for some reason sleepy or distracted.
Your argument is not rational.
Even though the non-sleepy eye might be much better than a good CCD Camera, many drivers would have hard quite a hard time to see the woman and react in time.
What counts are AVERAGE behaviour, not the 20% of drivers who only drive when fully healthy and with sufficient sleeping before.
We have lots of slightly sick, sleepy and bad-eyesight drivers on the road. Computers are never sick and sleepy (at least automotive grade computers and software).
So - on average - even the inferior sensors of autonomous cars might be good enough to make these cars safer per million of kms driven.
Having said that - this case must be investigated by experts in order to suggest improvements if possible. Which is also quite normal in engineering - we learn from accidents ! (That is only a problem for the lawyer sharks and their idiot customers)
As others suggested, the woman might have been homless and depressed thinking "I will just cross the road and if I am hit, so be it and my miserable life will be over".
That is of course a very sad explanation, but we all know this type of people exist, especially in the so-called "western world". Many people lack a family for various reasons, many of them quite egotistical. The downside of "freedom"...
I too have perfect sight in hindsight.
In which world do you live ?
A world with infinite financial resources to anyone ?
I work in the automotive industry and I can tell you that every 30-cent component is being discussed. For example, we measure voltages in one ECU, but we do not have a high-fancy voltage reference for our adc. Because that reference would have added something like 30 cents to a euro. Too expensive and not required for most applications of the ECU. We just use the voltage regulator of the MCU as reference.
If a car costs 30000 euros, then even 300 dollars for a great sensor will be a long discussion...
You describe the ideal situation of a young, excellent eyesight, fully healthy, fully awake and alert driver.
Back in the real world we have lots of middle aged and old drivers. Bad eyesight, sleep problems, exhaustion, slightly sick - DRIVING.
There are high odds a human driver would have run her over unbraked.
10 m is still more than 12 feet further than liquidpele's originally claimed 20 feet.
AC with imaginary PhD in Physics and a master debater.
If you are going to claim an education don't post AC. Otherwise you are just shoveling BS.
I think the public should know
1.) At which time (relative to impact) did any system/sensors detect the woman
2.) What was the decision of the software (levels)
3.) Which part of the system made the decision to ignore the detection.
That does not mean Uber is guilty or not guilty. All it means we should get to the bottom of this accident.
What sort of idiot crosses the street without watching for traffic???
An alert driver may have seen the woman however even if they didn't this is proof that having a driver behind the wheel "that can take over in the event of the self driving car coming across a situation that it can't handle" is a load of shit because by the time you've actually realised something has happened, you've been alerted to it and got over the initial shock and even thought about putting your hands on the controls the accident is already well in progress and you're just a passenger.
I only please one person per day. Today is not your day. Tomorrow isn't looking good either. - Scott Adams
As a driver of 10 years (without serious incident), I can pretty confidently say if I was driving that car in those conditions at that speed I would not have hit her (assuming I'm paying full attention which I always do while driving). There's enough time to swerve across to the left of her. Even if there was an oncoming car or another reason I couldn't leave the lane, there's certainly enough time to brake enough to make it an injury (e.g. broken leg/pelvis) and not a death.
A self-driving car correctly using LIDAR should have seen her way before I am able to, and therefore that accident shouldn't have happened.
You have to enable every fucking thing under the sun to get this video to play. I am NOT enabling googletagmanager.com, amazon-adsystem.com, or googletagservices.com, because I do not want to be systematically advertised to, nor do I want google to tag me so that they can track me.
Posting a link to a site like that is doing the advertisers' work for them. Fuck you, Slashdot. Fuck you twice. Here is a link to the guardian, which doesn't make you bend over for google and amazon just to watch a video.
"You're right," Fisheye says. "I should have set it on 'whip' or 'chop.'"
Another possibility -- the LIDAR profile of a pedestrian slowly walking a bike could have resembled a something that the computer was designed to ignore. For example, signs on the side of the road, or fences, would generally cause a lot of false signals unless the system was designed to ignore these.
If you look at the sign on the streetlight at the right, it appears to flash yellow twice around 1-3s into the video.
Does anyone know what that's about?
She was a crackhead stealing a bike:
https://arizona.arrests.org/search.php?fname=Elaine&lname=Herzberg&fpartial=True
Don't give a fuck if she got hit by a car. I bet theft, car break-ins and prostitution drops a bit in that area now that she is gone.
The he, is a she or at least identifies that way.
While humans need light, should the car be seeing with something better... like radar?
It could have seen the person and taken its course of action anyway, {...} It could have decided that swerving out the way was more dangerous, for instance.
Unless Uber have been implementing their driving system in a completely weird way (e.g.: the whole thing is just a giant deep neural net black box, with sensors on one side and controls directly driven by the output of the NN on the other side, like the Nvidia demo platform - which is just a techdemo, not a real-world use case), there are low-level system in drive assistance systems.
Even on current "Level 1" system currently on the street for quite some time, it the FCAS detects an obstacle in front of the car, the FCAS will hit the brakes and try to decelerate as much as possible (either stopping before hitting the object if distance permits, or in the current case if the obstacle jumps into view at the last moment, trying to shed as much kinetic energy before the impact as possible to reduce damage and potentially make the collision non-fatal).
Even if the higher level path-finding doesn't react (and frankly what should it do ? swerving is rarely a good idea as it increases risk of skiding), the low-level system should hit the brake.
Here they didn't, meaning that they didn't register the bike to begin with.
Somehow, not only the victim wasn't seen by the camera (obvious from the video : she was in dark until the last few seconds. Neither a human nor a camera would have noticed them in time), but it wasn't seen by either the lidar nor the radar.
Either she was invisible to the sensors (e.g.: the dark jacket aborbing the IR light used by the lidar, the spokes of the bike's wheels shattering the radar wave in a weird way, etc.)
or the sensors didn't register her (e.g.: the radar detects her as an object, but to the side (not in the same lane) and thus not a collision danger at that time. My experience with ACC is that sometime the radar has problem undestanding which object are in the same lane or not (problems fusing radar detections with the camera's lane detection and/or with the drivers wheel's heading) sometime missing (false negative) objects which are a bit on the side, or sometime getting confused (false positive) when the lane is curving a lot and the radar picks something in front of the car).
"Sufficiently advanced satire is indistinguishable from reality." - [Tips: 1DrYakQDKCQ6y52z6QbnkxHXAocMZJE61o ]
Jokes about stealthy bike aside, she *is* indeed very stealthy.
- Her jacket is very dark, it might be also dark in the IR spectrum and be as invisible on the LIDAR as on the camera.
- She not only doesn't have any light turned on on her bike, she doesn't even have any reflector on her bike nor on her clothing (here around where I live that would be considered suicidal)
On the other hand, the radar should have picked-up her (unless the bike wheel's spokes are scattering the radio wave in a weird way), but maybe the radar saw her too much to the left side, i.e.: "in a different lane and thus not on a collision course". The radar analysis part of the system should be upgraded to take into acount object moving perpenducaly to the vehicle.
(i.e.: Yes, if she stayed where she was, she would have been in a different lane. But she was moving side ways to the right. If the radar was able to pick her up, the computer should be able to calculate speed/trajectory and determine she'll be in the car's lane by the time the car reaches her).
"Sufficiently advanced satire is indistinguishable from reality." - [Tips: 1DrYakQDKCQ6y52z6QbnkxHXAocMZJE61o ]
How well does your cell phone camera in video mode work using the flash?
Why would you think they would spend extra cash getting a really great night time video camera to record thousands of hrs of video no one will ever watch?
At least it is better quality than the average security camera video, that is shot in a well lit store. I think those are all web cams from 2004, shoved into a new body.
But the amount you would decelerate in that 20 feet would have made the crash survivable.
I know this will be an unpopular opinion on Slashdot, but I feel the machine should always be accountable in these sorts of situations. I don't give a crap that she isn't crossing at a cross walk. Human life (even stupid ones) should be valued over machines and corporations. It doesn't matter that the street is dark. The car should have some sort of sensor on board that could detect her in sub-optimum lighting conditions.
We idiot proof most machines (chainsaws being one of the major exceptions off the top of my head). We should idiot proof automated cars. Its function is to get its passenger from point A to point B, not saw a tree in half. Safety should be a concern here. Don't act like you wouldn't have been grief-stricken if this was your idiot daughter who got smear across the pavement.
Due to the apparent total darkness where the bicyclist was crossing and the bright street lights right before, it seems that the camera was adjusting its DR to accommodate the near conditions. The human eye (not an old person) would have been able to see much further than what the video suggests.
She was 100% in her right.
What failed was the tech: inadequate senors, TOTAL PROGRAMMING FAILURE to even consider this situation, and that douche behind the wheel is to blame for her death as much as the tech and should be charged with manslaughter.
All you corporate shills, trying to evade a lawsuit (which is inevitable) are contemptible fvcking pigs for blaming her.
BTW, there's lots and lots and lots more of these kind of situations that will be happening as self-driving cars become more commonplace.
Self-driving cars a fvcking stupid idea, as in an ice or snow storm, you're fvcked, driving thru massive potholes, or incapable of recognizing nails in the road.
And I just can't wait for all the petty bitching about where it wants to park vs where you want to park in a parking lot.
Finally: autonomous vehicles wo/human occupant = theft magnet so huge that we will all die LOLing when car after car gets jacked up and striped or merely gets a brick thru its window. LOL
It's clear that a human driver would certainly not have seen the person before it was too late.. BUT you expect from a selfdriving car to be able to see into the dark using radar, so IT should have noticed the person much sooner. But it still might have been impossible to prevent the accident. Selfdriving cars will never be able to cope with all situations safely, but it will be much MUCH safer than with a real person behind the wheel..
It's going to be a long time before I let a driverless car drive me around. And I can't believe that these things are allowed on the road when they are clearly flawed. This is just as bad as when the Tesla ran full bore into into a giant trailer. The first, most basic thing a driverless car should do (safety wise) is be able to tell that something is directly in front of it. In both of these cases, it was completely unaware that there was an object in front of it. If your 'driverless' car can't do this most basic task, then it needs to go back to testing tracks and R&D. Not drive around on public roads...
This video looks doctored [to favour the potentially liable].It looks to me that it has been edited to have additional black/dark splotches to make the woman’s image seem darker than was actually the case in the original footage.On top of that, I believe that the human eye (such as of the human operator in the vehicle, had he been looking at the road) should have seen more than just the supposed utter blackness to the left.
I don’t have time to spare right now to extract frames from this video and include them here, but if you play it in a video player that lets you move frame by frame, here’s what you can discover and discern:
There is a pair of consecutive frames in the video (just before the collision) where the woman’s dark jacket is not illuminated in the earlier frame, and then illuminated in the immediately following frame.We can also see that the street lighting is orange in colour, whereas the vehicle headlights shine something distinctly lighter in colour than orange.Her head and hair/hat are also lit in orange several frames before this pair of frames.It seems very likely to me that blackness was added up to and including the earlier of the two frames, because the jacket should not just suddenly turn orange, it should become orange at the same time her head turns orange, which is several frames earlier.
Furthermore: Observe the first 2 seconds of the clip.Looking at the left side, you can see how much of the road, curb and even grass/dirt is illuminated by the headlights to the front-left of the vehicle.Estimate how far the headlights seem to be illuminating: the entire width of the adjacent lane, plus the curb, and more.
Now move the clip forward through the next seconds up to the collision.Observe how much that same front-left area of the vehicle is NOT lit anymore as it approaches the black splotches.The headlights should still be illuminating at least the full width of the adjacent lane.But they aren't.
In my opinion, this video was intentionally altered.
(This isn't even considering any LIDAR or radar that others have already mentioned.)
Uber may not want to bring their cars to Seattle any time soon.
Seattleites do this all the time -- just dart out into fast moving traffic because, damn it, pedestrians ALWAYS have the right of way (they don't legally -- Seattle has the same traffic regs as most U.S. cities -- but Seattleites are so insanely entitled and self-absorbed that the attitude has become the norm).
And that's just the pedestrians. The bicyclists are even more arrogant.
We have actual proof of teleportation! Its a conspiracy. The car industry did this on purpose. Oh and Im from the future so I know.
Regardless whether the victim made a bad choice about crossing or was oblivious to the vehicle. The problem is that the car did NOT DETECT A PEDESTRIAN RIGHT IN FRONT OF IT. This is not an edge case for testing. This is the primary test condition. How could it have failed so catastrophically in such a fundamental scenario? If the cameras couldn't see in the dark, what about the LIDAR and RADAR? I wonder what the data will show and I hope it is made public soon.