|
Owlofcreamcheese posted:If I’m so dumb why would I have more expert knowlage of on the video than someone that deals with car accidents professionally? Lacking expert knowledge has never stoped you before.
|
# ? Mar 22, 2018 04:29 |
|
|
# ? Jun 8, 2024 06:41 |
|
also, lets share a chuckle at the idea that the police chief deals with accidents on the reg
|
# ? Mar 22, 2018 04:33 |
|
BENGHAZI 2 posted:also, lets share a chuckle at the idea that the police chief deals with accidents on the reg And you do?
|
# ? Mar 22, 2018 04:35 |
|
Owlofcreamcheese posted:And you do? So let's recap: your arguments are A) appeal to authority, blindly B) "oh look at Mr smart guy over here" Oocc what specifically have the police said that leads you to believe Uber isn't at fault, other than "Uber likely isn't at fault"? Follow up, why are you blindly trusting that second thing with no proof to back it up?
|
# ? Mar 22, 2018 04:38 |
|
Solkanar512 posted:Serious question - why didn’t the LIDAR pick anything up? I'm not sure we (the general public) know if it did or not, or if the problem was elsewhere in the programming or a mechanical failure. We have the video, but not the computer's diagnostic data Uber would have, or the car's black box data the police have.
|
# ? Mar 22, 2018 04:44 |
|
Lol at the people still trying to defend their technology fetishism even when they see the person get killed.
|
# ? Mar 22, 2018 05:00 |
|
Doesn't this death mean less in the future? Even if Uber is 100% at fault, it's not as though this is a glitch that can't be fixed, especially now that Uber knows there's a problem with identification of pedestrians walking bicycles or whatever the problem is. Unless I'm misunderstanding how these automated cars work, it seems like these kinds of accidents, while tragic, are one of the things that most directly result in better safety in the future. It's is pretty gruesome that bug reports now involve actual humans dying, though.
|
# ? Mar 22, 2018 05:14 |
|
how long have they been open-road testing anyway, im kinda curious how long it took for a person to get killed in this case vs the original car back in 19-whatever
|
# ? Mar 22, 2018 05:22 |
IMO it'd be straightforward enough to do all the bug fixing without people dying, but it would be expensive and time consuming and seeing as society is now so married to profits and the idea that all new technology must be profitable from the very first version, that's not gonna happen. I understand why guns break Americans' brains but this bizarre goony fixation on the merits/evils of self driving cars I'll never understand and, if anything, the derails are even longer and more pointless and stupid. Are self driving cars any good atm? No, because there hasn't been enough time/effort invested because of the reasons above. Can they be perfect? Nah. Can they be better than a human driver? That possibility increases every day as people voluntarily set the bar lower and lower with their atrocious negligent driving whilst sensor and computing technology gets better and better. Can a human nannying a self driving car react like someone actually driving a car? gently caress no. We shouldn't even need driver nannies but again, we go back to the development vs profits equation. In the video in question it's pretty obvious the human couldn't do poo poo and the car just straight up ignored the woman's presence. OOCC is single handedly stoking the fires of the stupidest goddamn 'debate' on the internet with his mealy mouthed inane dumbfuckery and you stupid bastards keep responding to him like he's capable of reason. mandatory lesbian posted:how long have they been open-road testing anyway, im kinda curious how long it took for a person to get killed in this case vs the original car back in 19-whatever The very first automobiles were barely capable of human jogging speed, the biggest risk was fire and explosions. Slavvy fucked around with this message at 05:29 on Mar 22, 2018 |
|
# ? Mar 22, 2018 05:27 |
|
fire and explosions are just as capable of killing people as getting hit by the car itself
|
# ? Mar 22, 2018 05:31 |
Oh I thought you meant by collision. If we're opening it up to any mechanical misadventure I invite you to examine the state of OHS in 1899 and draw your own conclusions.
|
|
# ? Mar 22, 2018 05:34 |
|
Solkanar512 posted:Serious question - why didn’t the LIDAR pick anything up? I don't know how their system works but I imagine the bike reflecting light caused the sensor to think the object wasn't solid (possibly fog or rain?) I really can't tell, it seems like most computer vision systems I've seen would be able to tell there was an object in the path.
|
# ? Mar 22, 2018 06:04 |
|
JfishPirate posted:Doesn't this death mean less in the future? The way I see it, just based on the video I can imagine a normal driver having trouble seeing the person in the road. They're in between street lights and the angle of the headlights doesn't illuminate them until they're quite close. I'm sure the sensor on that camera is optimal in daylight so it's hard to say how dark it actually is in person. This exact scenario is where autonomous vehicles should be able to perform better than a human driver. mandatory lesbian posted:how long have they been open-road testing anyway, im kinda curious how long it took for a person to get killed in this case vs the original car back in 19-whatever I'm not sure about the public road testing for Uber, but it's been at least a year since Waymo and others have started operating. Most of these programs also have a number of "simulated" hours, so technically they should be dealing with an extremely large data set... but maybe Google is better at that than Uber. Slavvy posted:IMO it'd be straightforward enough to do all the bug fixing without people dying, but it would be expensive and time consuming I don't agree with this, the problem is that our roads, signals, and even the other drivers are too variable to have a simple rule set (or even a complicated large rule set). That's why a lot of these technologies are relying on machine learning to create models instead of trying to define a complicated rule set by hand. The problem with these models is that they use probabilities to generate a "score". There will always be a margin of error with these systems, that's why having multiple types of sensors/systems will greatly improve the confidence in these "scores". If these models are presented with a situation they don't see often the probability that it'll have a confident decision drops.
|
# ? Mar 22, 2018 06:17 |
Surely this:ElCondemn posted:
Is the way to fix this: ElCondemn posted:I don't agree with this, the problem is that our roads, signals, and even the other drivers are too variable to have a simple rule set (or even a complicated large rule set). That's why a lot of these technologies are relying on machine learning to create models instead of trying to define a complicated rule set by hand. The problem with these models is that they use probabilities to generate a "score". There will always be a margin of error with these systems, that's why having multiple types of sensors/systems will greatly improve the confidence in these "scores". If these models are presented with a situation they don't see often the probability that it'll have a confident decision drops.
|
|
# ? Mar 22, 2018 08:05 |
|
here's a vid someone posted showing the actual lighting conditions on the road the uber AV hit that lady https://www.youtube.com/watch?v=1XOVxSCG8u0&t=32s so even more indefensible. uber killed someone with their janky tech Condiv fucked around with this message at 08:25 on Mar 22, 2018 |
# ? Mar 22, 2018 08:21 |
|
BENGHAZI 2 posted:Hey look it's footage of the crash Wth, it looks almost like the headlights were off? Maybe it's the camera quality (which I'd expect to be a lot better than my eyes for something that's meant to drive the car) but in my car I'd expect to see the lady the moment she enters the road. And yeah how could have the LIDAR based detection completely ignored a full sized human moving in from the side like they is baffling
|
# ? Mar 22, 2018 09:20 |
|
Condiv posted:here's a vid someone posted showing the actual lighting conditions on the road the uber AV hit that lady Uber was probably eager to post their video because it's so dark that uninformed viewers might come to the conclusion that the lady just appeared out of nowhere. It also shows that the safety driver wasn't paying attention. ElCondemn posted:This exact scenario is where autonomous vehicles should be able to perform better than a human driver. Yes. Other than it being dark (do those cars even use visible spectrum or is it all LIDAR?), the conditions were ideal. The lady isn't moving fast, there are no obstacles which could hide her, the road is more or less straight, no traffic, weather is fine... Unless some extraordinary new information appears, Uber should absolutely be responsible for this. I'm afraid that they might have set up some legal protections that pin everything on the safety drivers, who are probably disposable minimum wage contractors.
|
# ? Mar 22, 2018 09:20 |
|
Condiv posted:here's a vid someone posted showing the actual lighting conditions on the road the uber AV hit that lady Yeah I'm pretty loving convinced Uber darkened the video they released in bad faith holy poo poo. Can somebody please explain to me this blob of insidious darkness covering up the woman's shoulders even though the car lights are already clearly lighting up her head? I also really, really refuse that Uber's static built in cameras (even if in form of an additional backup dashcam which this might be from) achieve much worse lighting conditions than that guy's shakey phone video. Teal fucked around with this message at 10:29 on Mar 22, 2018 |
# ? Mar 22, 2018 10:25 |
|
The head reflects light much better than the dark jacket on her upper torso. I doubt they selectively darkened the spot, if they did anything it was probably adjusting the overall exposure or maybe the camera was just messsed up to begin with.
|
# ? Mar 22, 2018 10:48 |
|
Condiv posted:here's a vid someone posted showing the actual lighting conditions on the road the uber AV hit that lady I'll have tou know this is basically the same as saying 9/11 was a controlled demolition
|
# ? Mar 22, 2018 10:52 |
|
Teal posted:Can somebody please explain to me this blob of insidious darkness covering up the woman's shoulders even though the car lights are already clearly lighting up her head? her coat is black.
|
# ? Mar 22, 2018 12:33 |
|
Yeah this is an enormous gently caress-up for Uber even if they're not legally liable. You shouldn't be testing on public roads if your detection is so bad that you can't spot a person with a bike slowly going in a straight line in front of your car. Like even if the classification of person/bike failed because it was a person with a bike + bags on the bike, "large object slowly moving in front of car's path" should still trigger a response. At the very least if you're at that early of a stage, you should have two people in the car like Waymo used to (safety driver + engineer), then you probably wouldn't have the same issue of the safety driver just looking at their phone. Cicero fucked around with this message at 12:43 on Mar 22, 2018 |
# ? Mar 22, 2018 12:38 |
|
Owlofcreamcheese posted:her coat is black. Cicero posted:Yeah this is an enormous gently caress-up for Uber even if they're not legally liable. You shouldn't be testing on public roads if your detection is so bad that you can't spot a person with a bike slowly going in a straight line in front of your car. Like even if the classification of person/bike failed because it was a person with a bike + bags on the bike, "large object slowly moving in front of car's path" should still trigger a response. Seriously, this poo poo isn’t difficult and many others use similar practices. But nooooooo, folks like Uber are too cool for school to dare ask anyone in any related industry for any best practices or anything like that.
|
# ? Mar 22, 2018 13:11 |
|
I don't get why they released the camera footage, but not what the LIDAR and the rest of whatever sensors the car has in-place actually were "seeing"/detecting. Was it a fault in the detection algorithm or in the communication of the alert to the central unit and eventually the brake system? Did it even log anything?
|
# ? Mar 22, 2018 13:22 |
|
AbstractNapper posted:I don't get why they released the camera footage, but not what the LIDAR and the rest of whatever sensors the car has in-place actually were "seeing"/detecting. Was it a fault in the detection algorithm or in the communication of the alert to the central unit and eventually the brake system? Did it even log anything? Because releasing anything but purposefully degraded and darkened worse-than-russian-dashcam footage would make it obvious their poo poo hosed up beyond doubt.
|
# ? Mar 22, 2018 13:49 |
|
AbstractNapper posted:I don't get why they released the camera footage, but not what the LIDAR and the rest of whatever sensors the car has in-place actually were "seeing"/detecting. Was it a fault in the detection algorithm or in the communication of the alert to the central unit and eventually the brake system? Did it even log anything? I mean, they released the video first because it's a human readable format. It's a lot easier to tweet out a video than like, an array of hundreds of thousands of [x,y,z] points. They might release a rendering from the lidar data later, but it's more steps to make that than just release a video that is viewable as is. (even the direct video has people complaining that it looks like raw video off a camera instead of an iphone video where the phone heavily processes the data to make it more viewable)
|
# ? Mar 22, 2018 13:51 |
|
Owlofcreamcheese posted:I mean, they released the video first because it's a human readable format. It's a lot easier to tweet out a video than like, an array of hundreds of thousands of [x,y,z] points. They might release a rendering from the lidar data later, but it's more steps to make that than just release a video that is viewable as is. All video is heavily processed to become human readable. Anything else you want to defend?
|
# ? Mar 22, 2018 14:00 |
|
Owlofcreamcheese posted:I mean, they released the video first because it's a human readable format. It's a lot easier to tweet out a video than like, an array of hundreds of thousands of [x,y,z] points. They might release a rendering from the lidar data later, but it's more steps to make that than just release a video that is viewable as is. lol the people working on self driving autonomous vehicles can’t map 3D data to a 2D video format That’s some high grade bullshit
|
# ? Mar 22, 2018 14:04 |
|
Solkanar512 posted:All video is heavily processed to become human readable. Let me guess, you are the guy that bought a 1000 dollar camera then exclusively keeps it in jpeg mode because you can't understand why the photo on the left (raw) is better to capture in than the photo on the right (jpg) ugg, look at that ugly blob of black, why don't they just take this picture on an iphone 6!
|
# ? Mar 22, 2018 14:08 |
|
Solkanar512 posted:All video is heavily processed to become human readable. Do the LIDAR sensors generate a visual recording by default?
|
# ? Mar 22, 2018 14:10 |
|
Condiv posted:lol the people working on self driving autonomous vehicles can’t map 3D data to a 2D video format They haven't said they won't or can't release lidar renderings, just they haven't done so as fast as they released video. Since video is a lot less steps to release.
|
# ? Mar 22, 2018 14:12 |
|
Owlofcreamcheese posted:Let me guess, you are the guy that bought a 1000 dollar camera then exclusively keeps it in jpeg mode because you can't understand why the photo on the left (raw) is better to capture in than the photo on the right (jpg) Keep on loving that chicken!
|
# ? Mar 22, 2018 14:42 |
|
Solkanar512 posted:Keep on loving that chicken! seriously "why is this sensor camera look so bad compared to this iphone video" is really really dumb.
|
# ? Mar 22, 2018 14:44 |
|
It will be interesting (for a small subset of people) where really automatic cars will have problems. I don't think will be in the actual driving thing. Is because our society is shaped by people with neurosis and narcissism and is not always logical. And then, you have the random person that have crazy ideas to exploit dumb mechanism. I could totally see somebody having 200 automated crash drive to the bottom of a quarry. Or somebody making a automatic cars "cheat" on their owner the weekends. Or all type of crazy things. In the end, I am pretty sure that we will limit the use of automated cars to well defined problems, like the roads between urban areas OR we will change how roads work to be better for automated cars (and worse for humans). Automated cars are stupid, but we can make the roads "smart" by embedding hint technologies to help these cars drive better.
|
# ? Mar 22, 2018 15:02 |
|
Tei posted:It will be interesting (for a small subset of people) where really automatic cars will have problems.
|
# ? Mar 22, 2018 15:07 |
|
Hey, I have a idea. Lets add "voice activation" to cars, so you say out loud "Hey car, I need you", and the car unpark and run where you are (in the entrance of the mall) to pick you. What could go wrong?
|
# ? Mar 22, 2018 15:30 |
|
Owlofcreamcheese posted:They haven't said they won't or can't release lidar renderings, just they haven't done so as fast as they released video. Since video is a lot less steps to release. it doesn't take a day or two days or three days to produce that rendering oocc
|
# ? Mar 22, 2018 15:31 |
|
Condiv posted:it doesn't take a day or two days or three days to produce that rendering oocc Render what? data clouds? depth maps? concentric circles? volumentic? false color? color overlay? Like they didn't say they aren't releasing something but it's not like they are just gonna dump the data in the free version of unity and export the first thing they can mash their hand into.
|
# ? Mar 22, 2018 16:17 |
|
Condiv posted:it doesn't take a day or two days or three days to produce that rendering oocc Producing anything that can be used in a legal proceeding takes way more time than you think it does. You check and double check everything. You make sure everything used to generate the filing is checked in and backed up so you can reproduce it. Any changes made in the program that generated the filing are documented, reviewed and explained.
|
# ? Mar 22, 2018 16:19 |
|
|
# ? Jun 8, 2024 06:41 |
|
I didn't bother to check the source or maths so feel free to dismiss this figure as bullshit but somebody calculated it and allegedly self driven cars now are statistically 40 times more lethal for pedestrians per distance traveled than driver driven cars
|
# ? Mar 22, 2018 16:24 |