Armed Polite Society
Main Forums => The Roundtable => Topic started by: griz on March 23, 2023, 04:09:36 PM
-
I didn't realize they had turned loose driverless cars to navigate on their own. I saw this article:
https://abc7news.com/cruise-ride-service-self-driving-cars-driverless-vehicles-san-francisco-storm/12993121/
where a couple of the cars got a little confused and ventured into a closed intersection. That led me to wonder, when one of these cars does damage to another vehicle, or even a person, who ends up being at fault? The software engineers. The executive who made the decision. The DMV politician that approved the experiment?
Anybody out there knowledgeable in this new venture into liability?
-
Well, driver-less transit isn't new - like a tram or metro train I would assume liability would fall on the owners. I imagine they may subjugate to the manufacturer if it can be shown to be a design or manufacturing issue.
-
Tesla's autopilot is programmed to disable and relinquish control of the vehicle to its human operator in the seconds before any event confuses the computer and would lead to a crash. This then absolves Tesla from liability since the human was in control at the time of the accident.
-
Well, driver-less transit isn't new - like a tram or metro train I would assume liability would fall on the owners. I imagine they may subjugate to the manufacturer if it can be shown to be a design or manufacturing issue.
Yea, I can see that as a default on a system that where the risks are pretty well known. Trains don't leave the tracks and you can generally fence off the right of way. Roads are different, they're full of people doing things that are far less predictable.
Tesla's autopilot is programmed to disable and relinquish control of the vehicle to its human operator in the seconds before any event confuses the computer and would lead to a crash. This then absolves Tesla from liability since the human was in control at the time of the accident.
But when there's no human operator at all the computer, much like a driver, has to make a decision as to what the least bad option is. They may just decide to drop anchor, turn on the four-way flashers, and call for a human, but it has to decide something.
I think what gets me is the computer could do everything according to its program, but still run over somebody*. That's why i wonder if the programmer could be at fault, but I would think the company overall would have to take the consequences.
*There was one case where a car (not a Tesla IIRC) that had a human driver who wasn't paying attention. The car didn't "see" a bike rider until too late and the woman on the bike was killed. I think the car gave control over to the driver when it was too late to do anything. Admittedly the situation might have turned out the same with a human paying a normal amount of attention since it was dark and she was crossing the road where nobody would expect, but such things will happen as we transition to automation.
-
I'm with the AMA* on these cars - they are NOT ready for prime time.
Testing has shown them to have problems with pedestrians, motorcycles and emergency vehicles. They need more development & refining before using the general public as test subjects.
*American Motorcycle Assoc.
ETA: https://americanmotorcyclist.com/cmc-study-adas
-
They are already statistically safer than a human driver. How many more orders of magnitude safer do you feel they need to get? What's the goal you or the AMA have for miles driven between accidents before they are ready for primetime?
Everytime I hear this conversation I can not get a quantifiable measure of "good enough ". If you are holding out for autonomous cars to have zero accidents, that's an unserious goal.
It should also be noted that the ADAS systems in the article Cliffh linked are not autonomous driving systems. They are the much more ubiquitous lane keep assist and auto-brake systems that are designed around the premise a human is still paying attention.
-
It should also be noted that the ADAS systems in the article Cliffh linked are not autonomous driving systems. They are the much more ubiquitous lane keep assist and auto-brake systems that are designed around the premise a human is still paying attention.
This is my understanding. It would be ridiculous to have an autopilot system that said, "Oh crap! I don't like this! Back to you driver!"
I recall reading that autopilot systems on helos are so good now, that rescue helos can fly into and hover in all kinds of situations that a human pilot could never handle.
-
To be clear, while I like tech toys and new technology I am also not convinced autonomous cars are 100% ready. Tellingly neither is Tesla, who is the farthest along and still 100% in beta testing .
But if we are going to have the conversation, we need to have a goal post. "Once self driving is "x" times safer than the overall human population we will introduce it slowly." or somesuch. Expecting self driving technology to be on hold till it's perfect and has 0 accidents is not only luddite stupi7, but will kill more people as we let humans continue to run each other over while we argue over a million or 1.2 million miles between accidents for robots.
Of the top of my head, once the robots are 5 or 10 times safer, statistically, then humans, I say let them drive. Yes a robot will end up running someone over, but it's still better than drunk Uncle Cletus ping ponging his Ram home every Friday.
-
They are already statistically safer than a human driver. How many more orders of magnitude safer do you feel they need to get? What's the goal you or the AMA have for miles driven between accidents before they are ready for primetime?
The problem is the stats are cherry-picked.
Tesla AutoPilot stats are measured in miles per incident, but only cherry picks miles when the AutoPilot is engaged. The AutoPilot is specified to only be used on the highway, where miles go by fast and cross traffic is minimized, and traffic cohesion is maximized. Human stats are measured across the spectrum: school zones, neighborhoods, urban centers, etc.
Also, take a look at the middle line here and it reveals another dimension to this info. The middle line is Tesla vehicle safety stats with AP disabled. Teslas, along with many new vehicles, have an enormous sensor suite in them to detect problems in blind spots or inattentiveness. A 1984 VW Rabbit does not. But the AP stats, and non-AP-but-sensor-equipped stats, cannot include old cars or people who cannot afford new cars. Generally, affluent people buy new cars and affluent people care for their vehicles better than poor people. Sometimes poor people are poor due to being young and lacking a work history that demands a high salary, but other times people are poor due to being of low value in the labor pool and having poor work ethic or life choices. This second bunch aren't exactly the best drivers either, and there is a financial means of discriminating between competent and incompetent drivers going on here.
-
Teslas seem popular in our vietnamese community. I wonder if they know something we don't.
I saw one pull out of a parking space at the market, and drive up to the front door to pick up its people.
-
Teslas seem popular in our vietnamese community. I wonder if they know something we don't.
I saw one pull out of a parking space at the market, and drive up to the front door to pick up its people.
That was a transformer.
-
That was a transformer.
More than meets the eye, huh?
-
They are already statistically safer than a human driver. How many more orders of magnitude safer do you feel they need to get? What's the goal you or the AMA have for miles driven between accidents before they are ready for primetime?
Everytime I hear this conversation I can not get a quantifiable measure of "good enough ". If you are holding out for autonomous cars to have zero accidents, that's an unserious goal.
It should also be noted that the ADAS systems in the article Cliffh linked are not autonomous driving systems. They are the much more ubiquitous lane keep assist and auto-brake systems that are designed around the premise a human is still paying attention.
Which car is safer than a human driver? I thought the few examples of driverless cars are experimental, not a production version driving on the street like a person does. The advances they have made are remarkable, but I didn't think any automaker has a completely autonomous car yet.
As to the "good enough" question, I want a system that can, without a driver or any sort of remote assistance, be able to pull out of a parking space at a big box store, and drive to a specified address, navigating a construction area with shifting lanes, and going through intersections with traffic lights, be able to pass a moped that's riding in the right hand part of the lane, and do all that better than a experienced, sober, human driver who isn't on their phone.
-
Skips the parking spot, but has construction zones, shifting traffic, intersections, and a dude on a scooter.
https://youtu.be/qLNk4Hem24c
-
Skips the parking spot, but has construction zones, shifting traffic, intersections, and a dude on a scooter.
https://youtu.be/qLNk4Hem24c
That's way better than I thought it could do. I noticed on the screen, which I'm assuming is what it "sees", that it was picking up every vehicle, including the parked ones, and even most of the pedestrians that were on the sidewalk. Pretty good, even though legally you can't get in the back seat and take a nap on the way home.
-
This is a shorter video of a shorter, less difficult drive, but the Chicoms have similar capabilities on their domestic Huawei chips. You can see that car is tracking the surroundings on the dash screen pretty well.
https://twitter.com/_mm85/status/1639290869626081280?s=20
If I had to guess I'd say over the road trucking, as a career, has less than 10 years left.
-
Skips the parking spot, but has construction zones, shifting traffic, intersections, and a dude on a scooter.
https://youtu.be/qLNk4Hem24c
Impressive, around the 1:50 mark it was even able to avoid someone on a scooter crossing against the light.
-
I'd thought about not responding 'cause I'd pull up references, "you" would pull up references, then we'd agree that we're both right (in our own minds). But, reading the latest AMA magazine lead to this:
https://sfist.com/2023/01/31/sf-regulators-have-had-it-with-self-driving-cruise-waymo-mishaps-ask-state-to-halt-expansion/
The same story, different source:
https://www.nbcnews.com/tech/tech-news/san-francisco-looks-hit-brakes-self-driving-cars-rcna66204
92 incidents in 7 months; running over fire hoses, trying to drive off from law enforcement, loosing communication with their servers & shutting down. Just the normal stuff that most drivers do.
Regarding the liability question, I'd say first it would be on the "driver", then owner & then the manufacturer, depending on the specific case & causes.
-
I don't think there's enough case law nationwide yet to determine who's liable if the autopilot fails or comes up short, damaging people or property.
As I see it drivers are held responsible for THEIR careless/inattentive/reckless driving, so the autopilot should be held responsible when it screws up.
Since inanimate objects can't be tried or fined, then the people who put them on the road should be. Starting with the highest level corporate executive who signed off on it, PERSONAL responsibility ought to be assigned to the people in the chain. INCLUDING potential criminal liability.
I'd also like to see PERSONAL liability exposure to the government officials/bureaucrats who approved automotive autopilots.
-
Since inanimate objects can't be tried or fined, then the people who put them on the road should be. Starting with the highest level corporate executive who signed off on it, PERSONAL responsibility ought to be assigned to the people in the chain. INCLUDING potential criminal liability.
I'd also like to see PERSONAL liability exposure to the government officials/bureaucrats who approved automotive autopilots.
That makes no sense. If an Amazon loading robot or an assembly machine at a GM plant hurts someone we don't go try to hang personal and/or criminal liability on the executives, robot manufacturers or OSHA folks that approved the use of automation.
-
That makes no sense. If an Amazon loading robot or an assembly machine at a GM plant hurts someone we don't go try to hang personal and/or criminal liability on the executives, robot manufacturers or OSHA folks that approved the use of automation.
That's not actually the case - a Boeing test pilot WAS criminally indicted by a grand jury for his part in problems with the 737 Max automated flight control system. (He was acquitted by a trial jury.) A VW executive who participated in an automotive software scheme to bypass emission tests was jailed. And various other individuals have been prosecuted for worksite safety violations that caused death/injury - a few older cases are listed here: https://www.pbs.org/wgbh/pages/frontline/shows/workplace/osha/referrals.html So personal liability CAN be attached.
Current law may not adequately address liability for new tech like automotive autopilots - I generally don't like "more regulation" but maybe this is an area that needs some.
-
Sure, but personal liability does not normally attach until you reach the level of negligence, or criminal conduct in the case of the VW guy. I remember the Boeing pilot and I thought his deal was he knowingly downplayed the differences between models to avoid a new type certification. That's also criminal conduct.
If we were to look at self driving code and find that a company knew there software would do something wrong and shipped it anyway to make a sales goal, then sure I could see personal and/or criminal liability.
I thought we were talking more about self driving accidents, that were either not foreseeable circumstances, or physics precludes avoiding the accident, or mechanical failure.
-
Actually, software issues are exactly what I'm talking about. There was one accident where a driver was letting his Tesla drive itself, and the trailer of a tractor trailer was stopped across the road. I think a driver would have recognized it for the immovable object it was, but the car didn't see anything close enough to the road and plowed right into it. The law in that case said it was the human driver's fault, but if it was truly driverless, I'm asking who is at fault. To me, that kind of thing IS foreseeable, and that's the problem. The car is always attentive, but sometimes doesn't recognize the rare but inevitable oddball situation.
-
Actually, software issues are exactly what I'm talking about. There was one accident where a driver was letting his Tesla drive itself, and the trailer of a tractor trailer was stopped across the road. I think a driver would have recognized it for the immovable object it was, but the car didn't see anything close enough to the road and plowed right into it. The law in that case said it was the human driver's fault, but if it was truly driverless, I'm asking who is at fault. To me, that kind of thing IS foreseeable, and that's the problem. The car is always attentive, but sometimes doesn't recognize the rare but inevitable oddball situation.
Incidents like that are why I say the autonomous cars are not ready for prime time.
-
Except that Tesla Full Self Driving is in beta, and despite the name "Autopilot " the software that guy was using specifically requires not only an attentive driver, but checks to make sure your hands are on the wheel. The software actively tries to make the driver pay attention. Using that incident in this fashion is like saying ABS systems aren't ready for prime time because someone on their phone hit something.
I think Tesla agrees with you that their FSD isn't quite ready, because it's still in beta, which is kinda the definition of "not quite finished"
But that begs the question I asked at the beginning of the thread (the first time I agreed with Cliffh). We will need a quantifiable measure of what is "ready". Griz had a decent response to that, but that needs to be expanded to how often? Once, twice, a million times in a row doing all those tasks? Saying they can never fail is u realistic, and I think likely to be dismissed out of hand. Americans have never in history required new tech to be 100% safe before adoption, and I don't see this being the first bit.
Take those cabs in SanFran. 92 incidents in 7 months is not in itself a useful data point. How many cars? How many miles driven? How many "incidents" of the same basic type (blocking traffic, broken down cars, not in the correct lanes, etc) did human driven cars have in that same time period? The company said "Cruise’s safety record is publicly reported and includes having driven millions of miles in an extremely complex urban environment with zero life-threatening injuries or fatalities." If that's true, (need to verify PR hack) how many traffic fatalities and life threatening injuries happened in SanFran in those 7 months? Which is safer, the robocab or the average California driver?
Isn't that data we should be looking at when we are talking about implementing this tech? Because it's coming one way or another. America is to in love with gadgets and tech to ignore this one.
I will say that my bet, without actually deep diving the data, is those AutoCabs, and Tesla's beta FSDv11.whatever are probably safer than the average driver. And I think we want them to be better than that before we turn them loose. As a swag, I think better than the 90th percentile or so of drivers on the road would be where we want to be. Then the ubiquity of auto drive will mean that most of the drunks, phone watchers and generally careless distracted drivers don't bother and let the robot do it, for an overall safer experience, but that's just a SWAG.
-
By the same token, " millions of miles in an extremely complex urban environment with zero life-threatening injuries or fatalities" is also not a useful data point. How many times in those million miles did the autopilot* kick out and default to the human driver? That's analogous to a driver falling asleep, yet they don't count that.
For what it's worth, I'm looking forward to a true automated system where I can get in the back seat and sleep on the way home, but we're not there. I think the tipping point will not be when the system is better than an average driver. The average is skewed by drunks, texters, nearly blind, and just plain bad drivers. It needs to be better than a sober, alert, skilled driver. and I wouldn't be surprised if it's better than halfway there.
-
The question of “who is liable for a collision of a full self driving vehicle” is likely to be: the insurance company of the owner of the vehicle whether that owner is individual or commercial. I’m guessing that the insurance companies will suss out the realities of which self driving vehicles are safer independent of and better than the developers of the technology themselves - much less the marketing wankers.
AZs notes on how the reported numbers may not tell the whole story are very relevant, but again insurance companies are going to figure that out quick. I bet rates for a given make and model would have to regularly adjust as updates are made to the driving algorithm.
Another interesting question will be who is guilty of non-accident traffic violations? Again it will probably be the owner of the vehicle, but that may result in some mildly interesting court cases where someone gets a ticket because their robocar ran a red light.
Maybe by that time we will already own nothing and be happy.
-
I'd like to see the ruling on an AV running a red light. Is the owner/"driver" really responsible if he's asleep in the back seat?
I wouldn't mind getting an extra 1/2 hour sleep on the way in to work. Having a reliable AV wouldn't be bad on those long drives to visit family. It'd cut down on my driving time - we could crash for a few hours while the car continued driving on.
I wonder if the AV's come standard with built in alarm clocks?
-
How long until self driving cars become so significantly better than human drivers that driving by hand will be outlawed?
-
How long until self driving cars become so significantly better than human drivers that driving by hand will be outlawed?
As soon as .gov requires that .gov can take over and remotely control the autopilot function.
-
Another interesting question will be who is guilty of non-accident traffic violations? Again it will probably be the owner of the vehicle, but that may result in some mildly interesting court cases where someone gets a ticket because their robocar ran a red light. .
I'd like to see the ruling on an AV running a red light. Is the owner/"driver" really responsible if he's asleep in the back seat
At least in Tampa, if my truck runs a red light camera, I'm responsible no matter who was driving. It's not a moving violation, but rather a civil infraction held against the registered owner of the vehicle.
I suspect AVs would be treated like that.