Slashdot Mirror
Police Chief: Uber Self-Driving Car 'Likely' Not At Fault In Fatal Crash (arstechnica.com)
An anonymous reader quotes a report from Ars Technica: The chief of the Tempe Police has told the San Francisco Chronicle that Uber is likely not responsible for the Sunday evening crash that killed 49-year-old pedestrian Elaine Herzberg. "I suspect preliminarily it appears that the Uber would likely not be at fault in this accident," said Chief Sylvia Moir. Herzberg was "pushing a bicycle laden with plastic shopping bags," according to the Chronicle's Carolyn Said, when she "abruptly walked from a center median into a lane of traffic." After viewing video captured by the Uber vehicle, Moir concluded that "it's very clear it would have been difficult to avoid this collision in any kind of mode (autonomous or human-driven) based on how she came from the shadows right into the roadway." Moir added that "it is dangerous to cross roadways in the evening hour when well-illuminated, managed crosswalks are available." The police said that the vehicle was traveling 38 miles per hour in a 35 mile-per-hour zone, according to the Chronicle -- though a Google Street View shot of the roadway taken last July shows a speed limit of 45 miles per hour along that stretch of road.
Why does it look like an sidewalk?
Much like a river kills the person jumping in it?
...gis sdrawkcab (usually not responding to ACs; don't bother posting as AC)
There will be a thorough investigation of the vehicle, the programming, all of the data and details. Even if it is decided that the victim acted imprudently, such accidents always (at least around here, unless it was the police involved) are fully investigated, and the driver is rarely exonerated from all blame, just the proximate causal fault.
Now, for you ignats who see class discrimination in the description that the victim was pushing a bicycle laden with shopping bags, a word; the police are the upper caste in these situations. Corporations will be prosecuted more often than police officers, and more often than reputable members of the community, IE, government. Or favored citizens. This is not new.
There was more than one factor leading to this tragedy, and if the end result is change in how these vehicles monitor their surroundings to have more time to analyze and react, excellent, and if the result is a recognition that even self-driving vehicles are unable to avoid such accidents, just as even skilled and careful human drivers are, well, then we've learned that self-driving does not equal infallible. That's important, and useful information.
deleting the extra space after periods so i can stay relevant, yeah.
In my community, we have great sidewalks, many crosswalks and all that needed to create a safe and walkable community. What do the pedestrians still do, you ask?
Walk out into traffic if it's more convenient. If a car hits them after taking reasonable measures to stop, they ought to be liable for all of the damage caused including to the vehicle and the driver's therapy if required.
My wife knew someone who killed a pedestrian who just walked out into traffic like this without thinking. Totally unavoidable. The "victim" was the driver, not the pedestrian because the driver was obeying the law and some stranger decided "fuck the traffic laws" and made her party to an accidental vehicular homicide.
The police chief needs to get some facts straight about the technology of autonomous vehicles work. LiDAR comes from LASERs. From the VEHICLE.
Unclear which "shadows" Chief Moir is talking about. Streetlights are but a one illumination source at play here.
Yes likely. Look, the guy who's seen the video says that a human driver probably wouldn't have averted the accident. You, who haven't seen the video, are only going on "generally." This incident isn't "general," it's very specific.
systemd is Roko's Basilisk.
The police have said that even reviewing the video the pedestrian is walking in shadow until they step out into the lane of traffic. A human driver texting on their smartphone in one hand and stuffing a double 1/4 pounder into their face with the other would have had a much slower reaction time. They likely wouldn't even have noticed they ran someone over until the next time they stopped for gas.
Humans can adjust to changing situations, they can also ready body language. Most people slow down when they see someone on the side of tge road looking like they are going to step out. An AI cant read that sort of thing. They can only react tl basic things presented to them.
Doesnt the car have sensors that could have detected the person and her bike with bags? Dont get me wrong it appears that the pedestrian was in the wrong but something should have been detected and the car should have done something to try and avoid the accident. Maybe these cars are not smart enough yet. Though once fully certified I would expect them to be better at driving then people. Something to work towards I guess.
Interesting series of tweets: https://twitter.com/EricPaulDe...
The median looks like it has fancy, inviting paths, but it also warns you not to use them. And the actual crossing is kind of daunting...
It is a rather bad design, but it does look dangerous in any case, so if I wanted to cross that way I would exercise extreme caution...
Violence is the last refuge of the incompetent. Polar Scope Align for iOS
Not only was this car speeding, but it did not recognise a road side hazard and drive by cautiously.
The police said the woman came "out of the shadows", it's likely that the car couldn't see the women before she stepped out, any more than a human could have (or did, since there was a human driver in the car too who said she had no idea anyone was there to step out).
Most self driving cars DO respond to anomalies by the side of the road and slow down or move over... but again, they have to be things that can actually be detected.
Also, 38MPH on a speedometer is within the margin of error of measurement that it was probably going the speed limit. All speedometers are set to read a bit high, in any case 35 vs 38MPH would not have made a difference to the pedestrian or ability to stop (and some people are saying the speed limit is actually 45, which makes way more sense with a divided median)..
"There is more worth loving than we have strength to love." - Brian Jay Stanley
In a microsecond, it will request the files of each person from Facebook, Google and the IRS and calculate the value of each person' life. If the one person is more important than the other four, those people will be dead before the car even hits them.
#DeleteFacebook
Humans can adjust to changing situations, they can also ready body language. Most people slow down when they see someone on the side of tge road looking like they are going to step out. An AI cant read that sort of thing. They can only react tl basic things presented to them.
My impression is that they detect and react to the actual physical posture and motion. But they can't read the person and tell if he appears drunk, high, mentally challenged or in some other way odd and likely to do odd things. It's a bit like the difference between a dog on a leash and a street dog with no leash, to a human they pose very different risks. But without programming in a ton of "human" logic they'll look just the same to a computer.
Live today, because you never know what tomorrow brings
You are overcomplicating it. If a car ends up in a complicated situation where it has to guess at how many will get killed, the answer is always to just brake. Get the amount of energy in the collision down, and who knows, some people just might survive. If not, too bad.
Squirrels don't count for the evaluation, you are obliged to not risk anything to avoid a squirrel.
Finally! A year of moderation! Ready for 2019?
Those are valid ethical questions that have to be answered by the AI and the programmers. If a car is forced to make a choice between killing a squirrel or killing a child, will it treat them the same?
This is not a "valid ethical question". It is just silly.
How will it choose if it has to decide between killing 4 people or 1?
Unlike most humans, the SDC will do the right thing.
But these rare corner cases are not that interesting, because they are ... rare. Far more common are accidents where the correct course of action is obvious: hit the brakes. And SDCs are FAR better at that. A typical human takes about 1.5 seconds to realize what is happening, move a foot to the brake, and start depressing it. An SDC can do it in less than 10 milliseconds. At 70mph, a car travels more than 150 feet in 1.5 seconds. The response time will be even worse if the human is not paying attention.
An AI cant read that sort of thing.
Why not?
Without the video itself we won't be able to tell for sure. However, from the streetview imagery, there's a number of places she could have emerged from that would've kept her her hidden behind bushes or trees. And I don't think she stopped at the side of the road at all. If she did, she would have had enough time to see the car that clearly was not slowing down for her.
Another key takeaway is that this scenario can now be analyzed and applied to millions of future situations. I just wish all the various autonomous driving companies were sharing their work.
What a coincidence that it just happens to be a SDC that just happens to kill a pedestrian!
Uh, yeah. 15 pedestrians are killed every day by human drivers in the US alone. What kind of simpleton freaks out about one pedestrian killed in the entire history of self-driving cars?
What reason did the Uber car have for going 38 in a 35 zone?
Surely the speed limit was lowered from 45 to 35 for a reason, probably for safety reasons.
Can the car not read road signs? It doesn't have the excuse of "I was watching the road, not my speedo" for a minor speeding offence. Did Uber fail to update the map data when the speed limits changed?
The risk of death being hit by a car below 30mph is relatively low. It increases rapidly as speed increases.
9% chance of death at 30mph.
50% chance of death at 40mph.
Starts reaching 100% fatal over 50mph.
There's a reasonable chance the woman, who may well have been in the wrong, would still be alive if the car was traveling at or below the 35mph limit.
source: https://nacto.org/docs/usdg/re...
There's another study that showed a reduction in speed by 5km/h would result in 30% fewer deaths. That happens to be how much the Uber car was over the limit.
http://humantransport.org/side...
No one is programming a car that way.
The first rule is to anticipate and slow down before anything could happen.
The second rule is to brake.
And the third is to stay on your lane. Except you have a spare lane going same direction.
Neither a programmer nor a car is deciding if it hits 2 3 4 or 1 person. If the thing in its lane is not going away, and the car has not stopped in front of it: it is hit. As simpel as that.
What is next is a realtime auction between the life insurance companies of the potential victims to determine who gets hit.
Run by AI bidding bots?
Cost free eBook I read (by iBook/Kobo/Amazon/ObookO/Gutenberg etc.): "The Green Odyssey" by Philip Jose Farmer.
Tire markes would mean that you braked to hard locking the tires, causing you to lose traction. A system breaking at optimal efficiency would and should not leave tire marks under any circumstance.
Were there any tire marks indicating the car tried to swerve or brake to even try to miss the pedestrian? Or did it just roll right over her like she wasn't even there?
The last time I hit a deer, it came out of the shadows at the side of the road, and stepped out just in front of me. I didn't mean to hit it, but it got inside my reaction time. I didn't leave swerve marks (well not much), I rolled right over the poor thing like it wasn't there. I don't hate deer. It wasn't callousness or indifference. I was just not fast enough to do anything.
A human driver, valuing the life of a fellow human being (or any other living creature for that matter) would at least try to not hit them, even if that meant swerving into a stationary object (parked car, lamppost, etc) or having to brake so hard they were struck from behind by another vehicle. Did any of these things happen?
It's not that the machine didn't care, but a human would. There was a licensed human driver sitting in the driver seat. They weren't fast enough to do any of these things either.
Or did it just digitally shrug and keep going?
No, you can see that the car is still there in the various photos.
The answer to these questions matters greatly. These so-called 'self driving cars', ...supposedly has a reaction time better than a human driver; sure doesn't look like it from here.
I have no trouble believing that a machine can react faster and often more reliably than me. It cannot however react in zero amount of time. We don't know how to do that. I am running wetware that has a million years of field testing behind it, and yet I am still not perfect. I do not expect the first generation self driving cars be infallible when I am not myself.
Given time they may become the safest drivers on the planet, but not yet, not right now. Besides, even if it reacted in some vanishingly small fraction of a second, it takes time and distance to bring a ton of rolling steel to a full stop.
Is usually around 5mph. It's difficult to keep a car at a rock solid 35mph, even for a computer. Changes in elevation can quickly alter your speed and religiously adjusting for it isn't even always the safest thing to do.
One of the hard lessons I had when driving is that if you slow down too much aggressive or stupid drivers will take that as a signal to go. My first accident was a t-bone where a girl hit me because she was trying to do a left into a busy road. I saw her start to move and put on my breaks. She saw me coming and did the same, but then saw me breaking and decided this somehow meant I was going to come to a complete stop in the middle of a busy street (the only option that would have stopped the accident by then). If I had not breaked she wouldn't have gone and the accident wouldn't have happened.
What I'm saying is there's such a thing as too much caution. Now, maybe if we can get the meatbags off the road that won't be true anymore.
Hi! I make Firefox Plug-ins. Check 'em out @ https://addons.mozilla.org/en-US/firefox/addon/youtube-mp3-podcaster/
Francis :).
Hi! I make Firefox Plug-ins. Check 'em out @ https://addons.mozilla.org/en-US/firefox/addon/youtube-mp3-podcaster/
How far do these cars look ahead? In defensive driving, they teach you to look WAY up the road. 150 yards back from the intersection, you are more likely to see people running onto the road than 5 yards from the intersection; it may just be a flash of them seen between vehicles up ahead. Are these cars properly watching as they pull up? They should have to submit high definition video from the moment the car starts to when it stops, from the perspective of a driver. If that person that it hits becomes visible at any time and the AI doesn't show any reaction in some way, "she ran out in front of the car" isn't good enough if you're only paying attention 20 feet before the intersection. I know some people don't notice these things, but a lot of people do and it prevents accidents. I would rather have autonomous cars be modeled after defensive driving techniques and I am concerned that they are not.
Laws are rules for the court, but merely a bottom bar to hit for life. Think beyond laws in your actions always.
Actually, most self driving cars don't simply brake. I was in my Tesla S in the leftmost lane of a highway - doing about 65 mph - with auto steer enabled. A car drifted into the lane from the right. The Tesla did brake but it also swerved onto the shoulder to avoid the other car.
This all happened so fast I didn't really have time to react until after the fact. ( It's sort of an interesting question if it knew there was a shoulder to drive onto or if it would have driven me into a ditch if there wasn't one. )
Clippy: "It looks like you've hit a cow. Would you like help writing BBQ invitations?"
If it is clearly the woman's fault, then produce the video for us all to see. Please blur the impact though. I just want to see for myself how much time before the woman entered the lane until impact. Simple, where is the video?
My brain is magic. It's such a common (and strong) reaction whenever anyone mentions automation or AI that it almost seems like some kind of instinct.
Maybe this has all happened before....
Why would a machine have trouble seeing someone moving out of a shadow, are they telling us that they didn't bother to use something as simple as infrared sensors for night vision to avoid a dependence on visible light?
Are you telling us you've seen the IR video? No? Is it not possible that the person was in a place where IR couldn't detect them against the background?
Or, you know what, you're exactly right. 9 years of testing, who knows how many man-hours developing and researching, around 10 million miles driven, but all of those people just plain forgot that IR was a thing. Yeah, you're probably right, anonymous genius.
"Our two-party system is like a bowl of shit looking at itself in a mirror." - Lewis Black
Less than 10% of these accidents are due to mechanical failure
This research indicates mechanical problems are about 2% of accidents.
Enigma
Indeed. Good safety engineering practices dictate to always remove system energy as fast as possible when you have no clear strategy. The examples given in the press of these "ethical dilemmas" are bogus. The car in question will not have enough information about the situation and can only a) bake hard without steering (as that could make matter much worse) or only minimal steering and b) tell any following cars what it is about to do. If there is a crash, it can c) directly alert emergency services. This is far, far better than what a human driver could do after his slow brain works out what is happening. Starting to brake that 0.5...1 second earlier an autonomous car will be able to alone will safe countless lives, and that is the comparison with a non-distracted competent driver.
Most ACs are not even worth the keystrokes to insult them. Be generically insulted by this and ignored otherwise.
Wow. We thought you were here to discuss the topic.
I'm here to discuss the topic with anyone who is interested in having a serious discussion. I'm not here to discuss the topic with a jackass who is predisposed to demonizing driverless cars and openly admits that he doesn't care about safety statistics.
Unfortunately at this point in self driving vehicle development it's also a meaningless statistic.
Statistics like that are averages, not absolutes. If something happens on average once every hundred million miles, that's no guarantee that it won't happen 4 times in the first 2 miles. The problem is that there are so few miles that have been driven by self driving vehicles, and so few incidents to date, that there's just not enough data yet to make ANY comparison to the safety of the average driver.
If anyone from either side of the argument states that self driving vehicles are more or less safe than the average driver, they are not basing that on any form of fact, just guesses. These vehicles have the POTENTIAL to be much safer than human drivers, but so far we don't know. It all boils down to whether better programming and sensors, and the impossibility of distraction are enough to make up for the lack of intuition, ease of pattern recognition, and versatility of a human driver.
Right now self driving cars can't handle adverse weather. They also can't handle anything unexpected, only things they've been programmed to deal with. But they handle those situations and those conditions extremely well. Over time they will handle more and more situations well, and fewer will fall in to the not handled category. I do truly believe that Self Driving Cars will one day replace human drivers and be far safer. I also don't think we're there yet, nor that we'll be there for quite a while yet. I don't think it's a problem of safety right now though, I think it's a problem of ability. There are just too many "corner cases" where self driving cars can't handle a given situation at all, that would prevent most people from being able to rely on one as their only means of transport without the ability to take over when the system can't figure out what to do next.
Can a self driving car tell if 2 people on the side of the road are drunk and wobbling all over the place, or if its 2 friends horsing around.
Sure. In either case, the pattern of movement would be different from 2 sober people purposely walking parallel to traffic, and the prudent thing to do is slow down. An ANN should have little problem learning those patterns. Most likely, this is already a solved problem, or considering the millions of miles driven, there would be more than zero avoidable pedestrian deaths by now.
Or as another poster said a stray dor or a dog on a leash.
That also seems like a relatively easy pattern for an ANN to learn.
Do you think the engineers designing these systems are stupid? They have libraries of millions of scenarios, both simulated and real recorded events, which they use for training and testing. It is unlikely that you are going to think of anything new that isn't already handled.
Or maybe the person wasn't doing anything that would appear to be a problem until they suddenly changed direction immediately in front of the car?
You can't have the vehicle assume that all people have a death wish and are likely to dive in front of the car at any moment. If you program it like that it will never be able to move if there are pedestrians anywhere nearby. You have to assume that the person will behave in a somewhat rational way or your car will never be able to actually get anywhere.
I had an incident a while back where I was driving on a residential road at fairly slow speed, there was a kid running all out on the sidewalk beside me, I was watching him. As I passed him, without looking, he made an abrupt 90 degree turn straight in front of my truck. I slammed on the brakes and barely stopped. Had he turned 1/4 second later I wouldn't have been able to stop in time, had he turned 1/2-1 second later the best computer wouldn't have been able to stop in time. But there was also no reason to stop or slow down until he'd already made the 90 degree turn, as it was a highly unlikely thing for him to do. It was illegal, it was dangerous, and it wasn't something you'd expect anyone to do. I thought about it a lot afterwards, and have many times been in similar situations but where the kid didn't make that 90 degree turn. There's just no way I can justify driving with the assumption that every person on the sidewalk, median, lawn, etc, could at any time make that abrupt turn in front of me. I'd never get anywhere, and I'd likely get in a different type of situation caused by the road rage from any driver behind me.
Not all collisions are preventable. They never will be, and no technology can ever prevent all collisions. What we can do is prevent all AVOIDABLE collisions, and doing that would save millions of lives. Is that not worth doing, even if a few UNAVOIDABLE collisions still remain?
We already know the answer. There was a human driver, he did nothing. We also have a second opinion from a person that is presumably a qualified driver, who reviewed the footage. He agreed that the human driver, and the AI driver, likely followed the appropriate course of action.
So we know EXACTLY what a human driver would do, AND what the AI would do. They were the same.
There are 210 million licensed drivers, so maybe comparing stats for more vehicles than there are drivers isn't the most intelligent analysis.
"Our two-party system is like a bowl of shit looking at itself in a mirror." - Lewis Black
Quickly making a sponge cake to lessen the impact damage?
That sounds more like an error, than planned action.
Tesla autopilot isn't designed for that kind of avoidance. It's a driving aid that keeps you in a lane and at a constant speed. It can only change lanes when the driver tells it to.
More likely the other car obscured its view of the road markings, and it suddenly thought it was way off the centre line for the lane and moved over. The only emergency action autopilot is designed to take is braking.
const int one = 65536; (Silvermoon, Texture.cs)
SJW, n: "Someone I don't like, and by the way I'm a fuckwit" - AC
Exactly. We have to drive with certain assumptions about how people will behave or no one would get anywhere. There are plenty of times making turns where you have to assume the other driver will slow down rather than plow right into you. Otherwise there would be traffic jams everywhere while everyone waits for the opportunity to turn without relying on the other driver reacting.
This posting is provided 'AS IS' without warranty of any kind, implied or otherwise.
38mph in a 35 zone as measured by the speedometer would presumably put it almost exactly the speed limit as all speedometers read slightly high. Beyond that there's reason to believe the actual speed limit in that location is 45 not 35, or that it is in the process of changing from 45 to 35 at that intersection. In any case, the car was not driving at any excessive speed. Additionally, we usually refer to people who drive at exactly the speed limit, or below the speed limit, by the term "obstructing traffic".
The chief of police did not think the speed was excessive, and he's reviewed the evidence. I'll take his word for it over some random slashdot poster who neither witnessed the incident, nor reviewed any of the evidence.
All evidence so far points to this collision being completely unavoidable, by even the best possible driver (Human or computer).