|
Thwomp posted:why was Uber technology allowed on the street if it couldn’t detect a pedestrian walking at a normal pace into the path of the vehicle? specifically because arizona is attempting to poach tech firms from california and so is being laxer than california regarding self driving car testing https://www.nytimes.com/2017/11/11/technology/arizona-tech-industry-favorite-self-driving-hub.html quote:PHOENIX — Three weeks into his new job as Arizona’s governor, Doug Ducey made a move that won over Silicon Valley and paved the way for his state to become a driverless car utopia. boner confessor fucked around with this message at 04:15 on Mar 22, 2018 |
# ? Mar 22, 2018 04:13 |
|
|
# ? May 14, 2024 10:01 |
|
As an owner of a dashcam and having reviewed footage after the fact sometimes, I can say with high certainty that you can see things with your eyes while driving at night significantly earlier than a dashcam can. Like, 1/4 mile earlier sometimes. Lidar/radar shouldn't have that limitation though. I'm only using it as a counterpoint to the argument based on 1.4 seconds.
|
# ? Mar 22, 2018 04:14 |
|
Thwomp posted:On top of all that, according to Ars Technica, the pedestrian is only visible (and that’s only her shoes) 1.4 seconds before impact. It’d require an attentive driver making a split second correct decision to avoid the same thing happening in a normal vehicle. Not blaming the victim but even in the video, you can see her until you’re nearly on top of her. The headlights didn’t reveal her until the last possible moment. The camera is being fooled by the bright streetlights - a real human driver, focusing on the road, has much better visibility into the shadows than what you can see on the video. And in general, if there's a dark obstacle in the road, you should be able to see it get lit up by your headlights, react to it, and safely stop before hitting it. If that's not the case, you're driving too fast and/or need to put your high beams on.
|
# ? Mar 22, 2018 04:15 |
|
boner confessor posted:the frequency of the beam doesn't matter The frequency of the beam absolutely matters when evaluating a statement like "bikes aren't very "solid" and don't reflect much." You can only "see" an object with light if the wavelength of radiation you're using is considerably smaller than the angle subtended by the object, a longer wave will essentially wrap around the object and keep going without reflecting. So yes, if you're pointing a radar beam at a bike, and the wavelength you're using is about the same as the size of the bike frame, the fact that the bike is mostly empty space and most of it is really just some narrow pipes, you're not going to get a lot of reflection. You need a wavelength on the order of the size of the members making up the frame. A police radar is sufficient for a bicycle. A lidar operating in the near-IR has a wavelength of about a micrometer. It will reflect off a bike wonderfully. quote:compared to currently unknowable factors such as the sampling rate (number of "dots") and the algorithm uber was using to detect objects relative to the background behind those objects Agreed that those are relevant to the question of why this car smacked into a lady with a bike. They aren't relevant to determining the near-infrared cross-section of a bicycle.
|
# ? Mar 22, 2018 04:16 |
|
shortspecialbus posted:As an owner of a dashcam and having reviewed footage after the fact sometimes, I can say with high certainty that you can see things with your eyes while driving at night significantly earlier than a dashcam can. Like, 1/4 mile earlier sometimes. agreed, if the driver is paying attention. which this driver wasn't, she was playing with her phone ...which is exactly why people want this technology so bad, so they can play with their phones while "driving" which is why i doubt this technology will make driving safer Phanatic posted:They aren't relevant to determining the near-infrared cross-section of a bicycle. you are so completely missing my point it's impressive. this isn't a stealth fighter, it's a bicycle being detected by a flashlight lol
|
# ? Mar 22, 2018 04:17 |
|
Facebook Aunt posted:A walking pace is the correct way for a pedestrian to cross the street. Running is wrong. Assuming the car was going the speed limit on that road at the very least she misjudged how much time she had to casually walk across the street with no crosswalk. I do agree a sober attentive driver would have seen her, as she went flying over their windshield. I think most people here would have smoked her, assuming that isn't a lovely night camera and we really have no way of knowing. It was dark as hell, she was wearing dark clothing, no lights. With normal headlights it seems like you'd have just enough time for your brain to comprehend the obstacle and hit the brakes just before she was on your hood. From the brief clip it did seem like there was minimal traffic so it would be realistic to have your brights on which could have made a big the difference. Either way a car with **dar that's on a public street should have caught it long before a human could.
|
# ? Mar 22, 2018 04:19 |
|
shortspecialbus posted:As an owner of a dashcam and having reviewed footage after the fact sometimes, I can say with high certainty that you can see things with your eyes while driving at night significantly earlier than a dashcam can. Like, 1/4 mile earlier sometimes. Have fun arguing that in court against "It was dark, she was wearing dark clothes."
|
# ? Mar 22, 2018 04:19 |
|
Be a little careful about the video as it can be misleading. The human visual system handles such large contrast better than most cameras. Think how terrible taking photos at night still is without a serious camera. How often can you see the subject just fine but the picture comes out as junk? Still doesn't excuse the fact that the sensors should have seen her as they are not dependant on lighting conditions. Comparing the system to human drivers is irrelevant distraction when it should be compared to its peers.
|
# ? Mar 22, 2018 04:19 |
|
bumming your scene posted:Have fun arguing that in court against "It was dark, she was wearing dark clothes." All I'm saying is that the video isn't representative of what an attentive driver would see. Nothing more, nothing less.
|
# ? Mar 22, 2018 04:24 |
|
oohhboy posted:Be a little careful about the video as it can be misleading. The human visual system handles such large contrast better than most cameras. Think how terrible taking photos at night still is without a serious camera. How often can you see the subject just fine but the picture comes out as junk? Yeah. True enough. The other sensors are supposed to make up for the shortfalls in camera technology. If they can't make the sensors detect at least as well as my busted rear end middle aged eyes, then those sensors have no business driving a car. Musk's stupid vacuum tunnel busses would at least only crash into each other.
|
# ? Mar 22, 2018 04:25 |
|
Facebook Aunt posted:Watching traffic run into her? When she started crossing the street that car would have been more than far enough away to see her and avoid her. A sober, attentive driver should have been able to see her. Personally, I don't start to cross the street if there is a car on the street which will hit me if its driver fails to take evasive action, because I don't assume that all the drivers are sober and attentive. In fact I know that a lot of them are drunk and on their phones. I don't think she's alone in assuming/hoping that people will slow down for her though. If I was driving that car at that speed I'm not sure that I could have avoided her either (assuming I could see as well as that camera), so I think she would have been equally dead if any one of roughly 50% of the population was driving that car as I don't imagine too many people slow down when it's dark based on the distance their headlights illuminate. I don't think she deserved to die either, neither do lots of the people we see dying in this thread, but like them she could have done with a healthy respect for the killing power of large objects. Of course it seems really lovely that the lidar couldn't pick her up at night, or that the car didn't do anything about that. On a related note, it seems like my new car's headlights don't illuminate much of the road ahead of me and I can't see very far at night, is this a new trend in cars or just mine?
|
# ? Mar 22, 2018 04:25 |
|
Buttcoin purse posted:
Headlights are adjustable. You can tilt them up and down. There is a standard of how far it is supposed to go. You are supposed to have a dealer install and adjust them. But obviously they messed yours up.
|
# ? Mar 22, 2018 04:29 |
|
boner confessor posted:you are so completely missing my point it's impressive. this isn't a stealth fighter, it's a bicycle being detected by a flashlight lol Look, you said a bicycle doesn't reflect much. That's just objectively wrong. Stealth fighter, bicycle, radar, optical, the laws of electromagnetics are the same in all cases and light works the same way. bumming your scene posted:Have fun arguing that in court against "It was dark, she was wearing dark clothes." That would be *extremely* relevant in the event you were in court because you hit a pedestrian. It might be enough to obviate the penalties entirely, assuming you weren't breaking any traffic laws when you smacked him. "It was dark, she was wearing dark clothes" might prevent you from being charged with anything in the first place. http://host.madison.com/wsj/traffic...bfd4944612.html http://katu.com/news/local/man-wearing-dark-clothing-walking-in-poorly-lit-area-struck-killed-by-car-in-corvallis
|
# ? Mar 22, 2018 04:31 |
|
Jabor posted:If that's true for you, you're a bad driver. 'too fast for the conditions' is meaningless when the stretch of road is not supposed to include pedestrians walking in front of you. The car wasn't speeding and intentionally halving your speed when it gets dark is not without its own risks when other drivers start overtaking you. The car obviously must be flawed either in its sensors or its programming dept., logs will tell in time what it actually 'saw'. The driver also hosed up for not watching the road, but that's common. It's not certain that she could have reacted in time anyway. The pedestrian made the costliest mistake, though I wonder what the alternatives to her shortcut were. Urban planning in many areas does not take pedestrians and cyclists into account at all and using the legal passageways forces you to take long-rear end detours.
|
# ? Mar 22, 2018 05:20 |
|
Buttcoin purse posted:On a related note, it seems like my new car's headlights don't illuminate much of the road ahead of me and I can't see very far at night, is this a new trend in cars or just mine? Make sure you aren’t running the daytime running lights by mistake. Every single night I drive home from work I see 2 or 3 cars with no headlights on but they have their fancy DRL on. Illuminates the road ahead slightly, dashboard is still all lit up but your back taillights are out and you can’t see poo poo.
|
# ? Mar 22, 2018 05:27 |
|
Nenonen posted:'The car wasn't speeding Please cite your sources as all the news I see suggests the car was speeding but not amazingly so. As one example: https://www.google.com/amp/s/www.telegraph.co.uk/technology/2018/03/20/ubers-self-driving-car-speeding-killed-pedestrian-police-reveal/amp/
|
# ? Mar 22, 2018 05:28 |
|
Measuring 38 when the actual speed is 35 is within the accepted margin of error for a speedometer.
|
# ? Mar 22, 2018 05:29 |
|
Nenonen posted:'too fast for the conditions' is meaningless when the stretch of road is not supposed to include pedestrians walking in front of you. The car wasn't speeding and intentionally halving your speed when it gets dark is not without its own risks when other drivers start overtaking you. So what's your plan when a piece of timber falls off the back of a truck five minutes before you drive through? Plow into it at full speed because you weren't expecting anything there? If the effective range of your headlights doesn't give it enough time to respond to an unexpected obstacle, you're either driving too fast, or you need to turn your headlights up.
|
# ? Mar 22, 2018 05:31 |
|
ulmont posted:Please cite your sources as all the news I see suggests the car was speeding but not amazingly so. As one example: there's some confusion on that https://www.nytimes.com/interactive/2018/03/20/us/self-driving-uber-pedestrian-killed.html quote:The vehicle was going about 40 miles an hour on a street with a 45-mile-an-hour speed limit google street view as of july 2017 has that area marked as 45 mph
|
# ? Mar 22, 2018 05:33 |
|
quote:“We don’t need redundant brakes & steering, or a fancy new car, we need better software,” then-Google engineer Anthony Levandowski wrote in an email to Larry Page in January 2016. “To get to that better software faster we should deploy the first 1000 cars asap. I don’t understand why we are not doing that. Part of our team seems to be afraid to ship.” Shortly thereafter, Levandowski would leave to found his own self-driving trucking company, which was quickly acquired by Uber.
|
# ? Mar 22, 2018 05:37 |
|
And now that they've hit a jaywalker, they know to update the software. How could they have predicted jaywalkers without this real world test?
|
# ? Mar 22, 2018 05:37 |
|
There are no forms of senses, electrical or human that would allow something to see an object the size of a person with a bicycle slowly crossing the path directly in front of it. Also using looking at your cellphone while driving and then mowing down a person is a situation that is impossible to account for and therefore should have no repercussions for the driver. I know in my state if i'm staring at my cellphone while driving for around 10 seconds with out looking at the road and I kill a person by running over them when they were directly in front of me then I would receive no repercussions. Furthermore over driving your headlights is also an unavoidable situation that has not yet been solved with modern technology, much in the same way that radar and lidar has never been used for determining ranges.
|
# ? Mar 22, 2018 05:56 |
|
Omg the visual spectrum alone gives you drat near enough time to react holy poo poo how did radar/lidar not catch that a billion miles away Lmao shorting Uber is right (even though they're not publicly traded) wtf hahaha
|
# ? Mar 22, 2018 06:26 |
|
Gunshow Poophole posted:Omg the visual spectrum alone gives you drat near enough time to react holy poo poo how did radar/lidar not catch that a billion miles away probably the lidar did detect her but she wasn't flagged as a pedestrian or an obstacle by the collision avoidance system lidar classification works kind of on the doppler effect, so vision would be better perpendicularly and get worse parallel to the axis of travel. by taking multiple lidar images and comparing them, you can try to determine if an object is moving or not, how big it is, which way it is going, etc. and probably none of those properties of the woman's movement were enough to signal the car to slow down
|
# ? Mar 22, 2018 06:29 |
|
I don't know anything about sensors but is there a reason a simple infrared / other non visible spectrum camera isn't integrated into the package that these systems use?
|
# ? Mar 22, 2018 06:33 |
|
Gunshow Poophole posted:I don't know anything about sensors but is there a reason a simple infrared / other non visible spectrum camera isn't integrated into the package that these systems use? probably super easy to get bad signals given how many other applications there are for infrared lighting. like you'd be picking up glare every night vision security camera out there, etc. you'd never really know when you suddenly drive into an area bathed in infrared light and your sensor goes blind. it's the same problem as visible light optical sensors except the driver can't tell when it's light or dark out
|
# ? Mar 22, 2018 06:41 |
|
https://www.youtube.com/watch?v=562yb0wifPo
|
# ? Mar 22, 2018 06:54 |
|
Thwomp posted:On top of all that, according to Ars Technica, the pedestrian is only visible (and that’s only her shoes) 1.4 seconds before impact. It’d require an attentive driver making a split second correct decision to avoid the same thing happening in a normal vehicle. Not blaming the victim but even in the video, you can see her until you’re nearly on top of her. The headlights didn’t reveal her until the last possible moment. Dashcams are often lovely pinhole lenses with sod all light gathering. Your eyes (assuming you haven't saturated the rods by, say, watching Harry Potter on your phone while driving/ been passed by a brodozer in AA search mode) will see significantly more than the camera. https://www.youtube.com/watch?v=8boWQvgo18U&t=296s Here's something that's not killbot related. I felt sure that there was a dedicated uber murdermachine thread up, but I can't find it now that I look.
|
# ? Mar 22, 2018 07:07 |
|
I just started as a longshoreman at the Los Angeles/ Long Beach ports, what you got for me OSHA thread?
|
# ? Mar 22, 2018 07:34 |
|
boner confessor posted:probably the lidar did detect her but she wasn't flagged as a pedestrian or an obstacle by the collision avoidance system In the 2-3 seconds before collision she was less than 20 degrees off-axis and within perhaps 200 feet. I cannot emphasize enough how incredibly negligent it is to build a system that cannot or does not recognize a human figure in that position and orientation 100% of the time regardless of lighting or weather.
|
# ? Mar 22, 2018 07:40 |
|
Who would have thought that a company infamous for cutting corners, dodging local laws and regulations, “contracting” literally anyone with a drivers license and car, and putting barely-tested autonomous vehicles that couldn’t even detect stoplights on the streets would have been involved in the first casualty in this field?
|
# ? Mar 22, 2018 07:46 |
|
Boner confessor is an idiot who knows nothing about LIDAR. Car's high level algorithms must suck as there is no way the sensors wouldn't detect her. Software engineers and management should be charged for vehicular manslaughter
|
# ? Mar 22, 2018 07:51 |
|
Ready for the autonomous vehicle derail to be over and to get back to OSHA hilarity.
|
# ? Mar 22, 2018 07:53 |
|
Splode posted:Software engineers and management should be charged for vehicular manslaughter Yeah, Tesla dodged a bullet when the guy got decapitated for watching Harry Potter. I don't think Uber will be so lucky here. Which is a good thing because gently caress Uber for many reasons.
|
# ? Mar 22, 2018 07:56 |
|
Ok I'll do my best. Recently at work I used a pair of tweezers as an automotive fuse. I was checking to see if I'd got the polarity correct on the charging cable for a Chinese electric scooter. If I'd been wrong it would've been very ugly but 50/50 is good odds!
|
# ? Mar 22, 2018 07:56 |
|
Today at the airport I squawked two different things on the plane I was going to fly! 1) the door lock wasn't working 2) the fuel measuring stick was for the wrong model of plane barely averted disaster I tell ya! (the fuel stick thing is actually a pretty big deal)
|
# ? Mar 22, 2018 07:59 |
|
boner confessor posted:
Legally, it's one thing. In practice, as a general rule, the vehicle with the most lugnuts has right of way. [quote="Phanatic" post=""4823326"] But if you set the speed limits accordingly, like traffic engineers say you should, then you wouldn't be able to write tickets. [/quote] Reminds me of Montana's short-lived "reasonale and prudent" speed limit. Joe Blow in a rusted-out '72 pick-em-up truck would be ticketed for doing 85, Mario Andretti in a Ferrari F50 can go as fast as he dares. Of course they did away with it and went back to regular speed limits because too many people complained. In knife/fingat chat, my uncle's right fisrt finger is permantly stuck straight out because he tried to take a knife away from somebody when he was 18. And to add OSHA icing to it, he once was building a deck and drove a screw through said paralyzed unfeeling finger, and only noticed when he reached to get the next screw. Edit: Jabor posted:And in general, if there's a dark obstacle in the road, you should be able to see it get lit up by your headlights, react to it, and safely stop before hitting it. If that's not the case, you're driving too fast and/or need to put your high beams on. When Brock Yates and Dan Gurney set the record for NY-LA in the second Cannonball, they maxed out at 172 mph. The car still had more to give, but it was at night and they were outrunning their high beams. Also, when the news interviewed him after the highly illegal race (looping back around to the first part of this post, it was conceived in protest of the 55mph national limit that had just gone into effect), Gurney said "at no point did we exceed 175mph" and they took it as a joke, because Dan was the sort that would say such a thing as a joke when he'd actually been doing 90 all the way. But they AVERAGED 90, including all the times they stopped for gas, driver changes, and hefty speeding tickets (another team, or maybe Yates' team a different year, got pulled over something like like five times in as many miles when they entered Ohio.). Everything about the Cannonball Baker Sea-to-Shining-Sea Memorial Trophy Dash was OSHA as gently caress, but they proved their point -- nobody was injured. Yates shut it down after four runnings because of increasing traffi on the interstates, but there have been unofficial attempts since, and at least one guy has made a documentary of breaking the record. Chillbro Baggins fucked around with this message at 08:22 on Mar 22, 2018 |
# ? Mar 22, 2018 08:03 |
|
Sagebrush posted:In the 2-3 seconds before collision she was less than 20 degrees off-axis and within perhaps 200 feet. I cannot emphasize enough how incredibly negligent it is to build a system that cannot or does not recognize a human figure in that position and orientation 100% of the time regardless of lighting or weather. the question is if uber is particularly negligent or if the entire concept of lidar-guided self driving cars is negligent and uber happened to be the first to reach that milestone Splode posted:Boner confessor is an idiot who knows nothing about LIDAR. i know more than you do, pal. sorry if you're jealous
|
# ? Mar 22, 2018 08:03 |
|
Sagebrush posted:Today at the airport I squawked two different things on the plane I was going to fly! Surely there's another measurement method for the fuel level? It's not just the stick right?
|
# ? Mar 22, 2018 08:03 |
|
|
# ? May 14, 2024 10:01 |
|
Splode posted:Surely there's another measurement method for the fuel level? It's not just the stick right? There are fuel gauges in the cockpit, but they're not very accurate, and as the fuel sloshes around and crossfeeds back and forth you'll get varying readings. Knowing that you have "half a tank" or whatever isn't enough anyway; you have to know the exact quantity to do weight and balance calculations and also obviously to calculate if you have enough range and reserve for the planned trip. The most reliable way of verifying how much is actually in the tanks is to climb up on the wing and use the measuring stick. I also know that if I stick my finger in the hole and I can just touch the surface of the liquid, I have about 12 gallons in that tank. Not sure what the FAA's opinion is of that method. Big airliners have fancy capacitance-based systems that can calculate the amount of loaded fuel to the nearest pound, which is usually all they need, but if the fuel gauge fails there are still dipsticks stored underneath the wings that the pilots will use to manually check the quantity. Sagebrush fucked around with this message at 08:12 on Mar 22, 2018 |
# ? Mar 22, 2018 08:09 |