Author Topic: Uber self-driving car hits and kills pedestrian  (Read 1796 times)

D.W.

  • Member
  • All Members
    • View Profile
Re: Uber self-driving car hits and kills pedestrian
« Reply #50 on: March 23, 2018, 11:23:40 AM »
It's cool that the military has managed to outsource all this R&D for the next generation of warfare tech!

Dodging pedestrians today, targeting insurgents tomorrow!

Seriati

  • Member
  • All Members
    • View Profile
Re: Uber self-driving car hits and kills pedestrian
« Reply #51 on: March 23, 2018, 01:15:16 PM »
Depends on the design - well designed monitoring for sensors should be quite robust.  A failed sensor should be something that could be immediately detected by a reasonably designed algorithm.

It's less failed, than malfunctioning that concerns me.  I agree failed should be readily apparent, but misaligned?  impaired?  Dropping intermittently?

What happens when a sensor fails, does the car become a brick, or does it continue because of the redundancies? 

Seriati

  • Member
  • All Members
    • View Profile
Re: Uber self-driving car hits and kills pedestrian
« Reply #52 on: March 23, 2018, 01:19:45 PM »
Maybe a lawyer could offer a more informed opinion about the level of legal culpability, but in my view if you do something illegal and dangerous you've just taken your life into your hands, and you'd better take care.

At least from one image I saw, there's a bike lane on the road right by where she was hit, that's actually in between the driving lane and the turning lane.  Does it change your view if she was hit less than 10 feet from a bike lane?  I mean heck, if her bike took a tumble she could have ended up in that spot.  Would the bike lane have prompted a human driver to be more alert for bikes? 

I know, I am much more aware of potential hazards on the streets where I know joggers insist on running in the roads rather than the sidewalks.

LetterRip

  • Member
  • All Members
    • View Profile
Re: Uber self-driving car hits and kills pedestrian
« Reply #53 on: March 23, 2018, 01:35:34 PM »
Depends on the design - well designed monitoring for sensors should be quite robust.  A failed sensor should be something that could be immediately detected by a reasonably designed algorithm.

It's less failed, than malfunctioning that concerns me.  I agree failed should be readily apparent, but misaligned?  impaired?  Dropping intermittently?

What happens when a sensor fails, does the car become a brick, or does it continue because of the redundancies?

For the video sensors they each overlap so every pixel is covered by two cameras.  There is also the lidar which will give full coverage.

If two sensors disagree by more than the noise threshold, than one of them has failed and the vehicle should come to a safe stop.

jasonr

  • Member
  • All Members
    • View Profile
Re: Uber self-driving car hits and kills pedestrian
« Reply #54 on: March 23, 2018, 10:00:34 PM »
Quote
I find it hard to reconcile those two videos.  If this is accurate, I'd be willing to believe that an alert human driver would not have made this mistake.

No way. The lady basically jaywalked across an expressway in the dark. Almost no one would have stopped in time at that speed.

Quote
I stipulate that human drivers make other mistakes that a machine would avoid.

I'm not sure there was a "mistake" to make, as I doubt most human drivers could have avoided the accident based on the video. But I'd suggest human drivers make this "mistake" fairly often, actually routinely. It's probably happened a few times literally as we have been typing this, lol.

Quote
I'm also concerned about diagnositics. While there are many conditions that can impact a human driver, tiredness, sickness, drunkness, the vast majority are detectable by a responsible person.  Tech issues are very difficult to deduce, particularly for a machine.  If a camera is inefficient, or has a arc error, how robust is the self diagnostic?

Human beings are terrible at self diagnostic too, trust me. Or they diagnose the problem, and just ignore it - eg: driving while intoxicated.



Seriati

  • Member
  • All Members
    • View Profile
Re: Uber self-driving car hits and kills pedestrian
« Reply #55 on: March 24, 2018, 11:42:50 AM »
Jason, just to confirm, did you watch the video LR linked to, or only the one the media released. If LR’s is correct there was plenty of light to see her from a fair distance away.

Fenring

  • Member
  • All Members
    • View Profile
Re: Uber self-driving car hits and kills pedestrian
« Reply #56 on: March 24, 2018, 02:02:36 PM »
Jason, just to confirm, did you watch the video LR linked to, or only the one the media released. If LR’s is correct there was plenty of light to see her from a fair distance away.

To be honest I have difficulty believing the accuracy of either video. The first is virtually pitch black, which seems extreme for a stretch of road like that with lights, so I would suspect video editing to darken the picture. But contrariwise the second video is all distorted with bright yellow light and crazy reflections on the road - hardly the real color of lighting you'll see on the highway, and so I equally suspect this video of being doctored to enhance lights beyond human sensory norms.

The truth of what it looked like is probably somewhere in between, but either way the person crossing was blatantly not even looking or reactive in any way to the oncoming car. That's a serious road, and I would be terrified to go across it either at night or during the day. If I had to I'd probably watch like a maniac and then sprint across just to make sure. And it's not like a deserted country road where it was bad luck that a car happened to be coming.

jasonr

  • Member
  • All Members
    • View Profile
Re: Uber self-driving car hits and kills pedestrian
« Reply #57 on: March 24, 2018, 02:19:07 PM »
Jason, just to confirm, did you watch the video LR linked to, or only the one the media released. If LR’s is correct there was plenty of light to see her from a fair distance away.

I watched both. But even taking into account the second video, I am skeptical that most drivers would have stopped in time.

The car was doing about 40 mph. At that speed, factoring in normal human reaction time (there is always going to be a delay from the moment the eyes perceive a danger to the moment the body reacts), you are looking at a total stopping distance of 120 feet. That's about 8 car lengths total stopping distance from the moment the driver perceives the threat to the moment the vehicle comes to 0 mph.

The perspective is a little hard to make out, but just eyeballing it, I'd say the lady becomes visible for the first time when she's just stepping into view (at around 4 seconds) and at that moment she's about 3-4 car lengths ahead of the vehicle.

Now let's assume the lighting is alot better and assume that the driver would have perceived her almost immediately when she stepped off the curb. That adds maybe another 4-7 car lengths leeway (assuming 1-2 seconds extra lead time), which admittedly gives the driver adequate leeway to stop, barely.

But there are two x-factors here: 1) The fact that it's dark; and 2) The nature of the roadway.

This isn't some crowded urban street; it looks to be basically a slow freeway with minimal or non existent sidewalks. No one would be expecting a pedestrian in that circumstance or looking out for one.

The chances of a driver noticing a pedestrian instantly at the moment she steps onto the roadway, in those circumstances, are remote. If this was broad daylight on an urban street with significant pedestrian traffic, it would be a much easier call to make.

Again, I will concede, having done the math and examining the video, some human drivers could have stopped for the lady in time - although much depends on luck and whether the driver's eyes happen to be in the right place at the right time.

If this were a lawsuit, I'd suggest that the driver of the car could be found 5%-10% liable, with the jaywalking pedestrian eating at least 90%.

TheDeamon

  • All Members
    • View Profile
Re: Uber self-driving car hits and kills pedestrian
« Reply #58 on: March 24, 2018, 04:22:52 PM »
Jason, just to confirm, did you watch the video LR linked to, or only the one the media released. If LR’s is correct there was plenty of light to see her from a fair distance away.

To be honest I have difficulty believing the accuracy of either video. The first is virtually pitch black, which seems extreme for a stretch of road like that with lights, so I would suspect video editing to darken the picture. But contrariwise the second video is all distorted with bright yellow light and crazy reflections on the road - hardly the real color of lighting you'll see on the highway, and so I equally suspect this video of being doctored to enhance lights beyond human sensory norms.

I don't find the initial dash-cam footage particularly unusual for a night-driving situation with many dashcams. They're absolutely horrible at night, and that has a lot to do with how camera optics work when paired against the "problem" that headlights present.

The second video was a bit brighter than I'd expect to encounter at night, but not outside the realm of roadways I've encountered driving over-the-road at night, in particular in urban areas. But I do think that video is probably on the other extreme with a higher gain than the human eye would have. (This is also further amplified by high/dynamic contrast screens many of us now use)

But I do think the Uber DashCam is a much more extreme outlier in all of this. There should have been some indication that he was driving under streetlights, and there were none, at all. While the second video shows a "splash"(bright spot) I'd more typically expect from a street light. So on a scale of 1 to 10, with the Uber being 1, and the other being a 10, I'd suspect "reality" is likely somewhere around 7 or 8, all things considered and probably closer to 8 than 7. Based on what that second video, she was likely crossing on the "Inside Edge" of the "light splash" from the right-hand street light.

The problem there is her night-vision should have been "very good," as she had been out in the dark for some time. The driver was busy looking at bright screens(and I think reading texts, or watching videos/movies -- because he kept smirking/smiling while looking down) so his night vision was likely poor.  Which cycles back to "The driver failed to intervene because he was distracted, and the specific nature of the 'distraction' also significantly impaired his night-vision."

I actually have issue with my GPS units at night because of that, they're too bright as far as I'm concerned when driving at night, and I have the brightness settings as low as they go. Further, that complaint is in regards to "ambient light" they throw off where I am not directly looking at them.

But going back to the non-impaired cyclist, she may have thought "it was bright enough to be seen" crossing where she did, but it obviously wasn't in this case with the distracted driver. So this goes back to "cross AT the street light" (where it is the brightest) not "near the street light." (And even further, failing to follow long-standing advice for bicyclists to "make sure they see you" in general, no side reflectors, dark clothing, dark colored bicycle, so on)

Quote
The truth of what it looked like is probably somewhere in between, but either way the person crossing was blatantly not even looking or reactive in any way to the oncoming car. That's a serious road, and I would be terrified to go across it either at night or during the day. If I had to I'd probably watch like a maniac and then sprint across just to make sure. And it's not like a deserted country road where it was bad luck that a car happened to be coming.

The Men in Black line about "People are stupid" seems to be relevant here. I can even go above and beyond the night-time in Dallas experience.

I worked as a railroad signal maintainer for a while, taking care of signalized railroad crossings. I watched first hand as two different vehicles(different crossings) failed to notice an activated railroad crossing with crossing gate arms down and properly operating. They didn't even slow down, they hit the gate arms at speed. And they didn't even have the excuse of "they were using their phones" at the time. Worse, the second time it was a pair of old ladies, and between the two of them, they failed to respond to the crossing being active. Maybe they were just that deeply engrossed in conversation with each other, but really? For further clarity, this wasn't even them "racing the gate" (saw that happen too), the gate had been down for some time before they were anywhere close to it.

If people can drive into active railroad crossing gates at speed even with the lights flashing and bell ringing, as I've seen happen in person, I sure as hell am not taking my chances with cars as a pedestrian on a public roadway, particularly when I'm entering a "Vehicular right-of-way" even if the laws say the Pedestrian takes precedence, I'm not counting on the vehicle stopping until they actually do.

TheDrake

  • All Members
    • View Profile
Re: Uber self-driving car hits and kills pedestrian
« Reply #59 on: March 25, 2018, 11:12:59 PM »
The more I think about this...

Über has the footage of the safety drivers. I doubt this was the first driver not playing attention. Did they ignore this previous data that this driver was not performing their job?

yossarian22c

  • Member
  • All Members
    • View Profile
Re: Uber self-driving car hits and kills pedestrian
« Reply #60 on: March 26, 2018, 08:11:24 AM »
The more I think about this...

Über has the footage of the safety drivers. I doubt this was the first driver not playing attention. Did they ignore this previous data that this driver was not performing their job?

Almost certainly. But I doubt anyone actually was reviewing footage of the safety driver unless there was an accident. Also in defense of the person this is an almost impossible task. The individual has to maintain the concentration and attention to a mundane task that they aren't performing for hours at a time. I can virtually guarantee that almost all of their safety drivers have lapses in concentration and attention long enough to cause accidents if they were actually controlling the car. Now one would hope that they would have some common sense and listen to books on tape instead of goofing around with their phone they way this driver was. But virtually everyone is going to have their attention lapse in this type of job at some point.

TheDrake

  • All Members
    • View Profile
Re: Uber self-driving car hits and kills pedestrian
« Reply #61 on: March 26, 2018, 09:16:44 AM »
I don't disagree, attention lapse is potentially inevitable. That's why Tesla has systems in place (and under constant improvement) to alert the driver when they are wandering off, though they are far from perfect.

This is pretty much the same position as a safety driver for Uber. It wouldn't take much in the way of computer vision to tell when the safety driver is not watching the road ahead, alert, and log the event. Another factor would be how long the driver is in the car without a break, and how long the break would be. The linked article says they are operating in eight hour shifts.

Other ideas are putting more sophisticated driver alertness technology in. Cadillac is apparently testing infrared cameras to judge head position - which I'm sure would get the privacy people into a frenzy about an always on camera pointed right at you in your car.

One could also have a mini-game competition for uber test drivers that provides incentives for paying attention. Click the button when you see a red car. Alexa playing I spy with the driver. Provide live feeds via periscope, and crowd source your oversight. You probably won't play Tetris if there are 9000 people watching your feed. Test reaction time with a light that simulates a dangerous situation and measures how quickly the driver gets their hands on the wheel or taps the brakes.

This Wired article has more

Of course any focus on safety drivers has to recognize that Waymo (Alphabet) started operating cars without safety drivers in Arizona 5 months ago. California is following suit, though it will require that a remote driver be able to pilot by remote control.

Over a year, Waymo drove 350,000 miles and human drivers intervened 63 times. For a person driving 30,000 miles per year, this would be about 3 times per year. I know I've made more mistakes than that.

There are 50 companies who have a license in California that have a license to test autonomous vehicles. I cringe to think about the level of technology for the bottom 10%, no matter how much I might trust the top 10%. Underfunded, way behind the competition, and sitting on a potential gold mine is a recipe for disaster with life or death at stake.

Seriati

  • Member
  • All Members
    • View Profile
Re: Uber self-driving car hits and kills pedestrian
« Reply #62 on: March 26, 2018, 03:47:00 PM »
I watched both. But even taking into account the second video, I am skeptical that most drivers would have stopped in time.

The car was doing about 40 mph. At that speed, factoring in normal human reaction time (there is always going to be a delay from the moment the eyes perceive a danger to the moment the body reacts), you are looking at a total stopping distance of 120 feet. That's about 8 car lengths total stopping distance from the moment the driver perceives the threat to the moment the vehicle comes to 0 mph.

I agree with the rough math here, what I disagree with is the idea that the threat wasn't visible (which probably means we have to agree to disagree). 

Quote
The perspective is a little hard to make out, but just eyeballing it, I'd say the lady becomes visible for the first time when she's just stepping into view (at around 4 seconds) and at that moment she's about 3-4 car lengths ahead of the vehicle.

But isn't that the point?  She didn't step out of a dimension door, she was there prior to the 4 second mark.  I routinely note pedestrians, during the day, from dozens of car lengths away, whether or not they are actually in the road way.

Even at night, and especially in areas with bike lanes and typical jogger routes, I'm actively looking for them. 

But just to give you an actual example, I dodged more than one deer in worse circumstances.  Granted sometimes needing a good bit of luck, where there truly wasn't light. 

Even if there wasn't enough light to avoid the accident, there was enough to begin to mitigate it.  A computer controlled car should have begun breaking and tried to swerve.  If the second video is correct, again, I think it's very likely a human would have avoided or at least begun to mitigate. 

Quote
The chances of a driver noticing a pedestrian instantly at the moment she steps onto the roadway, in those circumstances, are remote.

Again though, I'm not sure I understand the constraint?  I'd expect a driver to notice a pedestrian even on the side of the roadway, or on the side walk.

NobleHunter

  • Member
  • All Members
    • View Profile
Re: Uber self-driving car hits and kills pedestrian
« Reply #63 on: March 26, 2018, 04:31:14 PM »
A car that doesn't attempt to notice and categorize everything within reasonable proximity to the roadway is driving badly. Just because something isn't relevant to driving now doesn't mean it won't be in the next second.

TheDrake

  • All Members
    • View Profile
Re: Uber self-driving car hits and kills pedestrian
« Reply #64 on: March 26, 2018, 06:24:36 PM »
Again though, I'm not sure I understand the constraint?  I'd expect a driver to notice a pedestrian even on the side of the roadway, or on the side walk.

Have you been a pedestrian recently? Drivers tend to ignore pedestrians, even when they are turning on a red light in a downtown urban environment. I'd hope for a driver to do so, and I try to do so myself, but I wouldn't expect it. People who ride motorcycles know better than anyone to identify potential threats - everything from people flinging street-side doors into the flow traffic to cars failing to notice brake lights and crashing into the back of the vehicle in front.

The relevance of the constraint, to me, is whether the headline should be "Killer AI scourge of pedestrians" or "AI better than human drivers on average, but still needs to improve"

Fenring

  • Member
  • All Members
    • View Profile
Re: Uber self-driving car hits and kills pedestrian
« Reply #65 on: March 26, 2018, 09:47:23 PM »
The relevance of the constraint, to me, is whether the headline should be "Killer AI scourge of pedestrians" or "AI better than human drivers on average, but still needs to improve"

Maybe it should be "if you do something fatally stupid even the superior AI driver may not be able to save you from yourself."

jasonr

  • Member
  • All Members
    • View Profile
Re: Uber self-driving car hits and kills pedestrian
« Reply #66 on: March 28, 2018, 09:01:48 AM »
One point I forgot to mention: I am speculating that if the Uber was driving at 40 mph, then that was likely the actual posted speed limit in the area. I can't imagine Uber permitting its Ai drivers to speed.

But human drivers never drive the speed limit. So had a human been driving, one should assume she would have been doing at least 50 maybe more. That skews the math even further against a human stopping in time.

LetterRip

  • Member
  • All Members
    • View Profile
Re: Uber self-driving car hits and kills pedestrian
« Reply #67 on: March 28, 2018, 11:56:32 AM »
The posted speed limit was 45, Uber was driving 38.

jasonr

  • Member
  • All Members
    • View Profile
Re: Uber self-driving car hits and kills pedestrian
« Reply #68 on: March 28, 2018, 07:23:15 PM »
The posted speed limit was 45, Uber was driving 38.

Then it is safe to say the human drivers were in the 50-55 range - pretty near highway speeds then. I hate to speak ill of the dead but this lady really really acted unwisely.

yossarian22c

  • Member
  • All Members
    • View Profile
Re: Uber self-driving car hits and kills pedestrian
« Reply #69 on: March 28, 2018, 08:20:46 PM »
The posted speed limit was 45, Uber was driving 38.

Then it is safe to say the human drivers were in the 50-55 range - pretty near highway speeds then. I hate to speak ill of the dead but this lady really really acted unwisely.

Agreed. But I find it troubling that the Uber car didn't slow significantly before hitting her. I don't know if I would have stopped in time but I find it likely I would have gotten my foot to the brake in time to at lease slow my car if not stop it. Uber's road tests seem a little quick to me. I feel like Google probably has more resources and has been working on this for longer than Uber has been a company. Poorly programmed self-driving cars cashing in on the improved safety of their better programmed competitors may be one of the most dangerous aspects of the first few years of self driving cars.

LetterRip

  • Member
  • All Members
    • View Profile
Re: Uber self-driving car hits and kills pedestrian
« Reply #70 on: March 31, 2018, 11:49:56 AM »
A Tesla just crashed into a center divider with autopilot on.  Tesla claims he had been alerted for 5 seconds to take the wheel and had plenty of time to respond.  Of course the audible warning if it came at the end of the time period might have been too late.

Quote
In the moments before the collision, which occurred at 9:27 a.m. on Friday, March 23rd, Autopilot was engaged with the adaptive cruise control follow-distance set to minimum. The driver had received several visual and one audible hands-on warning earlier in the drive and the driver’s hands were not detected on the wheel for six seconds prior to the collision. The driver had about five seconds and 150 meters of unobstructed view of the concrete divider with the crushed crash attenuator, but the vehicle logs show that no action was taken.

The reason this crash was so severe is because the crash attenuator, a highway safety barrier which is designed to reduce the impact into a concrete lane divider, had been crushed in a prior accident without being replaced. We have never seen this level of damage to a Model X in any other crash.

https://www.tesla.com/blog/update-last-week%E2%80%99s-accident

While the Tesla autopilot is less sophisticated than self driving - this seems like something the autopiliot should have avoided either by swerving or more likely just slowing down.
« Last Edit: March 31, 2018, 11:52:38 AM by LetterRip »

TheDrake

  • All Members
    • View Profile
Re: Uber self-driving car hits and kills pedestrian
« Reply #71 on: March 31, 2018, 10:45:44 PM »
Tesla autopilot is very dumb. It is there to augment the driver. A violent maneuver like a swerve would kill drivers, not save them.

LetterRip

  • Member
  • All Members
    • View Profile
Re: Uber self-driving car hits and kills pedestrian
« Reply #72 on: April 01, 2018, 01:58:44 PM »
Interesting article at Wired. apparently car radars used for adaptive cruise control ignore all stationary stuff (road, signs, etc.) when they are at highway speeds because of too many false positives.

https://www.wired.com/story/tesla-autopilot-why-crash-radar/

jasonr

  • Member
  • All Members
    • View Profile
Re: Uber self-driving car hits and kills pedestrian
« Reply #73 on: April 03, 2018, 07:59:17 AM »
I don't really understand what the purpose of this "semi" autonomous driving system even is. Such a thing strikes me as inherently dangerous and ill-conceived.

I would never trust this technology or agree to be driven in a vehicle using it.
« Last Edit: April 03, 2018, 08:01:56 AM by jasonr »

Seriati

  • Member
  • All Members
    • View Profile
Re: Uber self-driving car hits and kills pedestrian
« Reply #74 on: April 03, 2018, 09:47:40 AM »
It seems to me that a semi-autonomous system would almost be the worst possible combination.  It takes many of the decisions out of the driver's hands almost lulling them into a low attention state and then can't handle the crisis that requires a quick reaction.  Wonder if they actually did any testing to see how delayed or rapid the reactions of an assisted driver are when compared to a non assisted driver.

NobleHunter

  • Member
  • All Members
    • View Profile
Re: Uber self-driving car hits and kills pedestrian
« Reply #75 on: April 03, 2018, 09:55:09 AM »
Anyone use adaptive cruise control and lane keeping for a commute? I've used it once or twice for long distance driving but it wasn't enough for any bad habits to develop. It certainly made medium and heavy traffic less stressful but I'm not sure it did good things to my ability to react to risks from vehicles not in front of me.

TheDeamon

  • All Members
    • View Profile
Re: Uber self-driving car hits and kills pedestrian
« Reply #76 on: April 03, 2018, 11:44:13 AM »
Anyone use adaptive cruise control and lane keeping for a commute? I've used it once or twice for long distance driving but it wasn't enough for any bad habits to develop. It certainly made medium and heavy traffic less stressful but I'm not sure it did good things to my ability to react to risks from vehicles not in front of me.

Have adaptive cruise in my Rig, have a love/hate relationship with it. Don't have lane keeping though.

The adaptive cruise/collision avoidance CAN lead to driver complacency. That said, it is a huge aid all the same, and probably saved me from one accident in the two years I've had it. (Vehicle in front of me slowed suddenly without signaling in advance, while I was checking my side mirrors. CA sounded an alarm and hit the brakes as my eyes "hit forward")

That said, the CA system scares the #$%& out of me on the occasions I have deliberately tried to trigger it(usually before a lane change), as it presumes perfect stopping conditions and is not supposed to directly intervene until that point has been reached. Which often means "scary close" before it does more than beep at you.

Adaptive Cruise is awesome in heavy/moderate traffic though. Only complaint there is sometimes it can be too good at what it does. Where you're cruising down the road and not watching the speedometer too closely(bad me!) and find yourself following a random vehicle for a few minutes until you notice you are getting passed more than usual... Check the speedometer and discover you are traveling 10 mph below your set speed.

Also, getting the readout reporting the speed of the vehicle in front of you can be fun. Seeing just how much lead other people have (or don't have) in their feet.

That said, my company doesn't use some of the more intrusive systems that are out there. We can override almost everything except the impending collision intervention which is brakes(and "jakes") only. And THAT can be its own kind of thing on slick roads as it hits the brakes, begins to induce a skid, then lets go of the brakes because it detected a skid, rinse and repeat as control is regained.

Grant

  • Member
  • All Members
    • View Profile
Re: Uber self-driving car hits and kills pedestrian
« Reply #77 on: April 27, 2018, 06:34:57 PM »
Yeah, as an engineer I'm already cringing at the idea that people's poor understanding of statistics/risk and the media's desire for sensationalism will combine to create a totally baseless and irrational fear of self-driving vehicles.

I wish they added to this media story "and here's how many people were killed that same day by idiot drivers, and here's the fatality rate of human drivers vs. AI per vehicle...".

Anything that slows down our progress towards fostering a broad public acceptance of AI drivers is essentially contributing to the deadly status quo.

Forgive me if I find this just a little bit humorous.  I too find that the public's poor understanding of risk to be disconcerting. Particularly when it comes to the Deadly Status Quo TM.

I'd like to start with the statistics for motor vehicle fatality rate in the United Sates, by year, compiled by the National Highway Traffic Safety Administration, which I have so wonderfully found easily accessible on Wikipedia.  https://en.wikipedia.org/wiki/Motor_vehicle_fatality_rate_in_U.S._by_year

Mmmmmmm.  Delicious.  The fatality rate per 100,000 population in 2016 was apparently 11.59.  A Total of 37,461 deaths.  Probably most of them kids in their teens or younger.  That's pretty bad until you realize that the number of deaths actually has been going down.  The peak year was apparently 1969.  Which is strange because it was the "best days of our life", and you'd think with every young person afraid of dying in Vietnam they'd be particularly safe on the road.  It's even more interesting when you consider that the fatality rate actually peaked in 1937, with 29.357 deaths per 100K of pop.  I really don't understand why anyone would have wanted to drive in 1937 considering how gawd awful dangerous it was.  They must have all been crazy. 

This of course doesn't cross reference the deaths with the amount of vehicles on the road or the total amount of miles driven.  Those numbers actually peaked in 1921, but that's because we don't have numbers for years prior.  Apparently in 1921 you had 24 fatalities per 100 million total miles driven.  In 2016 we are at 1.18.  Statistically speaking, from that data alone, I have a 1 in a million chance of dying for every mile I drive.  Absolutely terrifying.  This is why I very carefully log my mileage and will make sure I stop driving at 999,999 miles driven. 

These stats of course don't tell us the entire story.  They don't tell us how age reflects risk.  From my experience as a Paramedic working on the streets back in the summer of '99, I strongly suspect that the majority of vehicular fatalities are teenagers.  I also strongly suspect that the majority of those teenagers were not wearing seat-belts when they decided to become one with the asfault.  Not quite sure how much alcohol has an effect, but I'm sure the ladies of MADD would probably have strong opinions on it. 

Really, until I can make Twikkie my chauffer, I think we should ban teenagers from driving and hook up breathalizers to all ignitions. 

Of course, none of this can tell us how awesome and safe Ultron is going to be at driving us around.  Given how awesome Windows 8 and iTunes is, I have no misgivings at all.  I only hope I have the bandwidth and cash to handle all the updates.  I'm sure that no matter, Maximilian will be a better and safer driver than a drunk teenager not wearing their seatbelt, and you will probably see a dramatic decrease in vehicular fatalities.  That's as many as 37 thousand drunken teenagers without seatbelts that will live and grow old, be productive citizens, and have little children of their own. 

Obviously there will be some hiccups.  The strongest opposition for driver AI may not come from people without a proper sense of risk assessment (lord so many), but from the Teamsters.  Lord knows what will happen when the unions become involved.  I suspect that the rule will simply be that you have to keep a minder in the vehicle, despite the fact that the AI does all the driving.  This will open up new opportunities for, as has already been stated, to read books, watch reruns of West Wing on your mobile device, or, most likely, to look at porn.  I can only imagine the cost of the psychiatric counseling needed for the first responders of the future having seen that so rare vehicular fatality corpse with their hand around their member and a puzzled look of "what happened" on their face.  I imagine we will be able to replace them with ED209 as well. 

So I look forward to the bright future where great engineers and Bill Gates is able to save me from the terrors of the highway and free me from the tedium of actual experience by creating KARR.  Or KITT.  Whichever one was the good one.  But I'd like mine to have the voice of Scarlet Johansen.  Which will enhance my viewing pleasure. 

Of course, the thing I am most afraid of is the same thing I am most afraid for: the beautiful children themselves.  They are the future.  Let them lead the way.  Because when an entire generation abandons driving as work, only a select few dangerous citizens will remain who continue to drive themselves, probably on closed tracks, for pleasure.  The day will come when Little Suzie is given the gift of an antique Mazda Miata by her eccentric great-great-uncle Grant, who taught her to drive in the underground touring circuit when she was 12.  That will be the day that Suzie, who has the only drivable car in high school, becomes the coolest in skool.  I expect the parents of the male football players will be severely unhappy with her.  And when that happens, every kid will want a Mazda Miata, or Volkswagen Golf, or Oldsmobile Alero.  An entire generation will rebel against KITT, because like Ricky Bobby, they want to go fast, and they find they like being in control, and that control is better than anything on HBO or Pornhub, because it is real, it is here, and it is now, and will be an escape from the rest of their lives that is entirely controlled for their safety. 

Then of course, you will have to deal with the irrational fear.  Because of course, today more people are afraid of flying than driving in a car.  Of course, while flying is statistically MUCH safer than driving, that doesn't stop some people never wanting to get on an airplane, or even just being afraid of flying.  The answer of course is that flying brings with it a loss of control.  Joe Sixpack is at the mercy of the pilots.  Joe Sixpack doesn't understand the general hazards of flight or the proceedures to be safe.  He's just a passenger.  And while it's safer to fly than drive, that loss of control is terrifying.  When the plane or helicopter starts going down, there is nothing for him to do except put his heads between his legs and kiss his ass goodbye. 

It's going to be interesting when the first major *censored*storm mass fatality hits due to Johnny 5 going haywire during a solar flare.  Oh the press will have a field day.  There won't have been a real mass fatality in years and years since they outlawed the AR-15.  THE      FEAR     WILL     BE    YUUUUGE.  Congressional hearings with Apple and Google and Pornhub CEOs.  Oh sweet lord the irrational FEAR! 

I just want you all to know that I personally welcome our future masters and the engineers that bring us Tay and Tesla autopilot, and want them to know that I was always on their side. 


Fenring

  • Member
  • All Members
    • View Profile
Re: Uber self-driving car hits and kills pedestrian
« Reply #78 on: April 28, 2018, 04:22:37 AM »
That was an amusing satire piece, Grant. Personally I'm less concerned about vehicular safety and more interested in preventing incompetent people from screwing up the flow of traffic. I will be *so* happy when rubbernecking isn't a thing anymore. I'm personally done with one hour slowdowns resulting from those 1/200 people who inexplicably brake hard when joining a highway...when there's no merge and they retain their lane. I have no problem with giving up my "freedom" in exchange for hundreds of extra hours a year not stuck behind people texting while driving. I don't particularly see how "free" I am with giant pieces of metal in front and back of me for extended periods of time.


LetterRip

  • Member
  • All Members
    • View Profile
Re: Uber self-driving car hits and kills pedestrian
« Reply #80 on: May 04, 2018, 12:07:39 PM »
The ars story is remarkably poor quality for Timothy.

The actual report focused on the autosteer and showed a 40% reduction in accidents so his claim that we can't attributed the safety improvement to autosteer is bizarre - it says we can right in the report.  The automatic braking has been in all of the Tesla's so we know it wasn't responsible for the reduction in airbag deployments and yet Timothy is attributing the reduction the the automatic braking.

What the new claim by NHTSA is saying is that they having examined the entire autopilot system (which is true) therefore Tesla's claim that NHTSA has said autopilot improves safety is incorrect.

However, autosteer was the only part of the system that anyone had any questions about whether Tesla's are safer with or without it.  Everything else is obvious that they reduce accidents.

Fenring

  • Member
  • All Members
    • View Profile
Re: Uber self-driving car hits and kills pedestrian
« Reply #81 on: May 17, 2018, 12:03:34 PM »
Just randomly came across an article that addresses something I mentioned above:

https://www.vox.com/2014/11/24/7276027/traffic-jam

Quote
If you've ever driven on a highway, you've probably seen it happen. Traffic slows to a crawl, then stops entirely. Minutes later, it begins to move again, and then suddenly, you're moving at full speed.

[...]

"We're usually inclined to think that these must be caused by an individual driver," Seibold says. "But the models show that even if all drivers drive by the exact same rules, and no one does anything wrong, these waves can still arise."

These jams, in essence, emerge whenever you have enough humans driving cars on a highway. So the only real way to eliminate them probably involves handing the wheel over to something other than a human driver.

This is one of the things I would like to see reduced with AI driving. Now this depends to an extent on the AI involved. Independent AI (for each car) will eliminate unnecessary braking and stupid mistakes - the kind of thing an idiot would do as a result of not paying attention, panic, incompetence, etc. However even that won't stop what's still observed in simulations when cars are treated as mere particles that all follow identical behavior. To entirely stop that would require a centralized, coordinated AI system that actually controls all cars on the road. In such a system at a green light (assuming lights were needed) all cars could literally begin moving at the same time at nearly the same speed (to create buffers between them).

But putting aside that kind of advanced system and focusing on independent AI per car, the simulations seem to show that some fluctuations occur and that 'phantom traffic' can still occur. But here's the thing: a minor disturbance caused by application of brakes or changing lanes (the two great factors in traffic) will certainly create a backwards-flowing shockwave that can even bring traffic to a halt at a certain point if there isn't enough buffering between cars, which in turn makes this possibility dependent on traffic density. But the important thing to remember is that these events will pass - meaning the shockwave will move backwards until it 'exits the system' and everyone can re-buffer and resume driving at normal speed. It's a one-time event that goes away, as the article states maybe lasting a few minutes as a result of some random confluence of factors requiring a significant brake. It can happen again at some random interval, and then pass again. So periodic slowdowns would still happen with AI, but wouldn't last all that long from the perspective of an individual driver because after applying brakes the cars could speed up again and resume normal operation, and the cars behind (after minor delay) could follow suit, etc etc.

The real problem occurs when one of two things happens: (1) Someone does something incredibly stupid, jamming on the brakes for no reason or whatever else, and this causes not just a shockwave but a massive one, and not just that, but the 'morale' of all drivers behind gets killed along with their velocity and they now accept that "there's heavy traffic" and adopt a slave mentality which prevents emerging from it - even if they get to the point where there are no longer cars in front of them blocking them, and yet they won't speed up. So basically the shockwave is worse and the re-buffering doesn't work properly. Neither of these scenarios would happen with AI - no 'stupid' errors, and no morale to break. (2) People are repeatedly making errors, which - even if the system is re-buffering correctly - inserts additional shockwaves over and over, in effect causing the traffic disturbance to be recreated so frequently that effectively nothing changes and the gridlock remains static. If sufficient numbers of drivers are making these errors, or even driving too slowly and causing people behind them to brake, then the traffic will become 'permanent' until the traffic density becomes so low that even these errors don't cause a significant disturbance. Even then you will sometimes see a "traffic pile-up" no more than a dozen cars long where they literally come to a complete stop (!!) because one person was too slow or something similar, except the event passes in seconds and they begin to go again so most drivers will never see this particular event. The worse part of (2) is that if the situation is bad then the mindset of the drivers will tend to devolve into "every man for himself" (which doesn't take much provocation even in 'civilized' society) and people will begin to initiate very grabby maneuvers where they gain minute amounts of traction with lane changes and so forth at the general expense of the fluidity of traffic flow. Basically it turns into Lord of the Flies and causes further degeneration of the system.

For both of these scenarios, which in my view are the chief causes of "rush hour" traffic slowdowns as we know it, even independent AI would eliminate them entirely (in theory). In practice the quality of the AI would be relevant so initial tests to this effect may or may not achieve the 'correct' result of what AI should be able to do, but you have to start somewhere. I can't wait.

TheDrake

  • All Members
    • View Profile
Re: Uber self-driving car hits and kills pedestrian
« Reply #82 on: May 17, 2018, 12:34:59 PM »
Interesting. I wonder if net throughput is truly affected? The solutions described seem to describe anticipating slow downs and, well, slowing down. So there are no waves, but you still don't arrive sooner?

Either way, you don't need full on automation, just traffic-jam assistants or stop-and-go adaptive cruise control

This is a dated article, it has been around for a while. We don't need new technology, we need to get people to buy new cars with better features. Unless you'd be willing to go for a DOT regulation? :D

Fenring

  • Member
  • All Members
    • View Profile
Re: Uber self-driving car hits and kills pedestrian
« Reply #83 on: May 17, 2018, 12:51:01 PM »
Yeah, the article is dated, but it's not like these kinds of analysis are using cutting-edge maths. I mean, ok, fluid dynamics is still in its infancy, but running simulations of basic particle-behavior in 2-3 lanes doesn't require chaos theory, just brute forcing it by running the simulations a zillion times.

And I think that while adaptive cruise control would certainly be a pleasure when making a long drive, I'm not sure how effective it will be in mitigating the sorts of things causing major traffic slowdowns. These things tend to involve the 'human factor' and as long as that's still in the picture - deciding when/how to change lanes, whether to signal, whether to step on the brake or not, etc - then the problem won't go away. Many times people are applying the brake for literally no reason; they just sort of drive with a foot over the brake at all times and apply it as a matter of course, even when no one in front of them is slowing down. Maybe they went 1 mph over the limited and wanted to slow down, who knows. There are people who get agitated when a slow driver is in front of them or when they're in traffic, and will sharply accelerate towards the car in front of them and then brake sharply again, just to vent steam or amuse themselves. This move alone causes significant problems! As does rushing towards the car in front of you when there's a gap and then braking when you arrive at the desired distance, as the cars behind you think that's going to be the cruising speed and don't know that you're just bridging the gap. None of this would be helped much by adaptive cruise control, assuming the drivers take a hands-on approach and apply the brakes all the time anyhow. Nothing will stop these shenanigans other than removing the human from the equation altogether, IMO.

TheDrake

  • All Members
    • View Profile
Re: Uber self-driving car hits and kills pedestrian
« Reply #84 on: May 24, 2018, 07:02:19 PM »
Quote
At 1.3 seconds before the crash, the self-driving system realized emergency braking was needed to prevent a crash. But Uber had disabled the feature to reduce the potential for unwanted braking, such as for a plastic bag in the road.

"The most shocking portion of the report is emergency braking maneuvers were not enabled," said Bryan Reimer, research scientist in the MIT AgeLab and the associate director of the New England University Transportation Center at MIT. "Is the driver expected to look at the outside world continually? It's impossible when you're providing tasks that interfere with that."

oops.

I've come almost full circle on this one. That's just a negligent decision by program management or engineering. Emergency braking is one of the few things that are well proven to avoid or mitigate accidents. If your emergency braking is broken, then stay on the test track until you get that figured out. :(

yossarian22c

  • Member
  • All Members
    • View Profile
Re: Uber self-driving car hits and kills pedestrian
« Reply #85 on: May 24, 2018, 09:49:58 PM »
Quote
At 1.3 seconds before the crash, the self-driving system realized emergency braking was needed to prevent a crash. But Uber had disabled the feature to reduce the potential for unwanted braking, such as for a plastic bag in the road.

"The most shocking portion of the report is emergency braking maneuvers were not enabled," said Bryan Reimer, research scientist in the MIT AgeLab and the associate director of the New England University Transportation Center at MIT. "Is the driver expected to look at the outside world continually? It's impossible when you're providing tasks that interfere with that."

oops.

I've come almost full circle on this one. That's just a negligent decision by program management or engineering. Emergency braking is one of the few things that are well proven to avoid or mitigate accidents. If your emergency braking is broken, then stay on the test track until you get that figured out. :(

So Uber is a company full of misogynistic a**holes who treat their drivers like dirt and make negligent decisions about testing their auto driving software. Why is this company still allowed to exist? I hope this woman's family sues them for everything they are worth.