|
Paradoxish posted:Unless I'm misunderstanding you, you seem to be talking about two fundamentally different things. Cheesus is asking how self driving cars should deal with situations where the posted speed limit is 30mph, but every single human car is driving 50mph. He's not talking about the car understanding the posted speed limit. i posted earlier that dealing with the posted rules/social rules dichotomy is super difficult for automated vehicles. but when it comes to "the speed limit changes every quarter mile" that would be relatively easy for a robot car to deal with, as we already have GPS devices that can warn drivers when speed limits are about to change which is how i'm interpreting "calvinball driving rules" from this anecdote Cheesus posted:Her: You can't go that slow here! of course this would mean vehicles that annoy human drivers by strictly following posted speed limits
|
# ? Jun 13, 2019 15:21 |
|
|
# ? Jun 8, 2024 15:27 |
|
There Bias Two posted:Even with the issue of randomly crashing into 18 wheelers, I'm not sure whether these kinds of incidents make them less safe on the whole than a manually driven car. Is there enough data out there to examine this yet? There's not enough data, and more to the point, the data is meaningless for ascertaining the level of safety. You see, all 'driverless' cars still have a human operator who can take over. Therefore, a ratio of accidents per 100k miles or whatever is a meaningless metric - how many accidents were prevented by a human taking over? Back in the very early days, they did release those numbers of course. They were barely lower than the global ratio, which is tabulated from all kinds of roads. In comparison, autopilot or whatever it was only drives on the safest, easiest roads, which makes it look better than it actually is. Of course, then the ratio of accidents grew so they probably can't even claim that anymore.
|
# ? Jun 13, 2019 15:22 |
|
The speed limits in North America should be comprehensively reviewed, revised and then strictly enforced. That's not really a driverless-car issue, it has much wider implications.
|
# ? Jun 13, 2019 15:24 |
|
Has anyone crunched the numbers on how the Model S compares to other large luxury sedans like the E class, S class, and A8 in deaths and injuries per mile driven?
|
# ? Jun 13, 2019 15:25 |
|
PT6A posted:The speed limits in North America should be comprehensively reviewed, revised and then strictly enforced. That's not really a driverless-car issue, it has much wider implications. No, but it's a problem that directly affects driverless cars. It's also just a smaller subset of a larger class of problems involving social road rules. Plus there's effectively no chance of this kind of massive speed limit reform happening any time soon, and I don't think driverless cars are going to change that.
|
# ? Jun 13, 2019 15:29 |
|
PT6A posted:The speed limits in North America should be comprehensively reviewed, revised and then strictly enforced. That's not really a driverless-car issue, it has much wider implications. it'll probably cause a lot of limits to go up in places in the US, they lowered the speed limit below the recommended speed limits on most highways because of the 70s gas crisis.
|
# ? Jun 13, 2019 15:32 |
|
Paradoxish posted:No, but it's a problem that directly affects driverless cars. It's also just a smaller subset of a larger class of problems involving social road rules. Yeah, definitely agreed. It's a huge problem for driverless cars, but it's not just a problem for driverless cars. It's really a problem for everyone. From what I see driving, no one actually pays attention to the speed limit. They'll just go whatever speed they generally feel like going, whether that's above the speed limit or below it, and it's contrary to the interests of road safety because you have a bunch of people just doing their own thing with a stunning lack of situational awareness.
|
# ? Jun 13, 2019 15:35 |
|
gently caress MA and MA drivers. I set the cruise control 2 under the limit and sit in the right lane because I've got out of state plates and MA staties are the worst people in America. Massachusetts cops just make up laws on the spot and dare you to sue them.
|
# ? Jun 13, 2019 15:37 |
|
OJ MIST 2 THE DICK posted:it'll probably cause a lot of limits to go up in places in the US, they lowered the speed limit below the recommended speed limits on most highways because of the 70s gas crisis. Yeah, in Canada too. On a decent day, the flow of traffic can be 130 km/h on limited access highways, and there's nothing inherently unsafe about that speed. At the same time, you can't really say "well, they should make the speed limit 130 km/h" because then given the North American mindset, people will try to push it to 150-160 km/h and at that point it becomes unsafe. Highway speed limits should definitely go up, but it has to be accompanied by a change in driving habits and enforcement that says "hey, we've given you a new, higher limit, but it's actually a limit now and we expect you to not exceed it."
|
# ? Jun 13, 2019 15:38 |
|
MickeyFinn posted:Oh, please. The problem you keep running in to here is that you don't have anything to say on any of these topics that isn't about your fantasies about what technology will be like in X years or that technological advancement is inevitable, so no one should question it. I don't know what to do with your posts on the subject. Hell, it might even be interesting to read an explanation about why you seem to be so enamored with technology. Certainly more so than another few vacuous paragraphs about how great technology is. On the other hand, I can think about specific criticisms of technologies and I can reject them, accept them, argue about them or forget them. What? I started by saying it looks like a lot of people putting money into it seem to think it will come about, maybe they're not wrong since a bunch of progress has happened over the last decade, and all I got in response was people saying I've got crazy tech fantasies or it wouldn't even be desirable if someone did start selling a working driverless car. As for the perfectly myopic 'is progress actually good tho?' schtick I'm not even going to be drawn.
|
# ? Jun 13, 2019 18:18 |
|
Surprise Giraffe posted:What? I started by saying it looks like a lot of people putting money into it seem to think it will come about, maybe they're not wrong since a bunch of progress has happened over the last decade, and all I got in response was people saying I've got crazy tech fantasies or it wouldn't even be desirable if someone did start selling a working driverless car. As for the perfectly myopic 'is progress actually good tho?' schtick I'm not even going to be drawn. this is mostly the weird old man scared luddite thread.
|
# ? Jun 13, 2019 18:27 |
|
Surprise Giraffe posted:What? I started by saying it looks like a lot of people putting money into it seem to think it will come about, maybe they're not wrong since a bunch of progress has happened over the last decade, and all I got in response was people saying I've got crazy tech fantasies or it wouldn't even be desirable if someone did start selling a working driverless car. As for the perfectly myopic 'is progress actually good tho?' schtick I'm not even going to be drawn. people put a lot of money into trying to find the hole at the top of the earth that enters into the hollow realm, as well Mr. Fall Down Terror fucked around with this message at 18:37 on Jun 13, 2019 |
# ? Jun 13, 2019 18:35 |
|
PT6A posted:Yeah, in Canada too. On a decent day, the flow of traffic can be 130 km/h on limited access highways, and there's nothing inherently unsafe about that speed. At the same time, you can't really say "well, they should make the speed limit 130 km/h" because then given the North American mindset, people will try to push it to 150-160 km/h and at that point it becomes unsafe. Highway speed limits should definitely go up, but it has to be accompanied by a change in driving habits and enforcement that says "hey, we've given you a new, higher limit, but it's actually a limit now and we expect you to not exceed it." I think conventional traffic engineer wisdom is the speed limit should be the 85th percentile. So you should expect 15% of all drivers to break the speed limit. But if 99% of all drivers are breaking the written limit, then the written limit is too low.
|
# ? Jun 13, 2019 19:23 |
|
I mean from what I've experienced driving across the US, when I got to the 80 mph speed limit straight-level-with-slightest-of-curves freeways through nowhere? I ended up hanging around the upper 70s for speed and very few people seemed to be going over it beyond passing other vehicles. Like trying to get around truck convoys puttering along at 65-70 mph. I'm sure if I ever get on those Texan 85 mph limit freeways in their empty areas it will be about the same, with even fewer people consistently breaking that limit. And on the other side of things, you regularly end up cruising at 80-90 mph on the big Northeast Turnpikes on the clear sections when traffic is free flowing and with minimal enforcement. Particularly everytime I've driven through the original Pennsylvania Turnpike segment between Pittsburgh and Carlisle, it's real easy to creep up to 90 when you're not in the tunnels. Advertising for the road back before World War II advised travelers that they should slow to 90 for the curves, though surely most people's cars back then weren't keen to get up that fast.
|
# ? Jun 13, 2019 19:40 |
|
golden bubble posted:I think conventional traffic engineer wisdom is the speed limit should be the 85th percentile. So you should expect 15% of all drivers to break the speed limit. But if 99% of all drivers are breaking the written limit, then the written limit is too low. I always assume it's the limitation of having physical signs. Like I assume a traffic engineer could drive down a road and tell you moment to moment way more realistic speed limits pretty accurately, but is stuck boiling that down to one unchanging sign every such and such miles makes it wrong for any individual point at any individual time. Like they aren't going to put a sign up every 15 feet then change it every night or anything.
|
# ? Jun 13, 2019 20:08 |
|
golden bubble posted:I think conventional traffic engineer wisdom is the speed limit should be the 85th percentile. So you should expect 15% of all drivers to break the speed limit. But if 99% of all drivers are breaking the written limit, then the written limit is too low. theres a large chunk of the US where this isnt true because the roads were built to higher limits pre-Energy Crisis and were capped at 55 and never raised back after the law was repealed
|
# ? Jun 13, 2019 20:16 |
|
Xarn posted:loving lol at the guy who thought that if he pretended to be on his phone, people would be less angry with him In the California driver’s handbook it mentions that it is not your job to enforce the speed limit, and going the limit in the fast lane is an act of vigilantism. Mostly because the laws are written, as mentioned, to allow arbitrary stops, and that if you do that in LA you will literally be murdered.
|
# ? Jun 13, 2019 20:23 |
|
Surprise Giraffe posted:What? I started by saying it looks like a lot of people putting money into it seem to think it will come about, maybe they're not wrong since a bunch of progress has happened over the last decade, and all I got in response was people saying I've got crazy tech fantasies or it wouldn't even be desirable if someone did start selling a working driverless car. As for the perfectly myopic 'is progress actually good tho?' schtick I'm not even going to be drawn. See what you're missing is that to a large extent we're disagreeing with you that "a bunch of progress has happened" in self-driving. Everything they've done so far has been the basic groundwork that allows you to even begin the hard parts of the problem, which is also the sexy part because it's the part where you take a normal car and make it able to usually stay between the lines and only sometimes kill someone. The parts where you try and turn it into a real product that you can commercialize are both harder and much less sexy so there's no reason to think they'll somehow happen way faster. e: And just in general self-driving is a weird tech race, because unless you're Uber and are promising that self-driving will somehow turn your company around so please give us money, there is incentive to take your time and make sure you've got it right. Any benefit you get out of being first to market is obliterated if your first self-driving model ends up killing a bunch of people. Feinne fucked around with this message at 20:37 on Jun 13, 2019 |
# ? Jun 13, 2019 20:34 |
|
huh in the plains US our highway speed limits are usually 70-75 with some 65, maybe 55 at a weird bendy interchange, and driving at abt 80 is normal- 20 over would be ludicrous- but 20 under is about normal for snow/ice conditions worst conditions I've driven in, that an AI might actually be better at, was morning rush hour in San Antonio winter- fog reduced visibility to maybe two car-lengths, and commuters were whippin past my out of state rear end at 10 over while I'm driving 10 under anyway didn't know there was still a lot of driving 55 out east, that's bananas, but I guess you're a lot more tightly packed out there
|
# ? Jun 13, 2019 20:40 |
|
It's limits at 55, not driving at 55.
|
# ? Jun 13, 2019 20:45 |
|
There Bias Two posted:Even with the issue of randomly crashing into 18 wheelers, I'm not sure whether these kinds of incidents make them less safe on the whole than a manually driven car. Is there enough data out there to examine this yet? dex_sda posted:There's not enough data, and more to the point, the data is meaningless for ascertaining the level of safety. You see, all 'driverless' cars still have a human operator who can take over. Therefore, a ratio of accidents per 100k miles or whatever is a meaningless metric - how many accidents were prevented by a human taking over? I wanted to take a stab at a very simple version of the problem. So, I assume that the 1.16 fatalities per 100 million VMT (in 2017) are all separate incidents and comparable and so on. I further assume that the accidents are Poisson-distributed. With those (probably very flawed, but also extremely simple) assumptions in hand, I use human driving as the null hypothesis and find that if self-driving cars drove 250 million miles and never killed anyone, there'd only be a 5.5% chance the result was just a "lucky" sampling of regular-old human drivers. If you have 800 vehicles driving an average of 35 mph, doing that 24/7/365, you could do it in a year. It turns out human drivers are actually really good, which is why the usual use case for driverless cars is drunks. Also, considering that driverless cars have already killed two people that I know of, it is not looking good.
|
# ? Jun 13, 2019 20:48 |
|
https://twitter.com/verge/status/1139259231885025280
|
# ? Jun 13, 2019 20:55 |
|
MickeyFinn posted:
This runs contrary to everything I've ever heard about human driving performance.
|
# ? Jun 13, 2019 21:26 |
|
There Bias Two posted:This runs contrary to everything I've ever heard about human driving performance. humans are generally pretty good drivers so long as they aren't distracted. a huge number of car accidents are from teens, young adult men, old people, and folks who are doing things they shouldn't be doing like eating, speeding, racing, or playing with phone, or intoxicated/impaired. experienced drivers who are attentive and driving at a reasonable speed generally don't get in accidents - this is why insurance companies can hand out discounts for people who go multiple years at a time without getting in accidents (at least ones that are too bad to be ignored or forgiven by the other driver)
|
# ? Jun 13, 2019 21:41 |
|
Cheesus posted:If CT is worse than that, I can't even imagine how self-driving cars can possibly be programmed for New England Calvinball driving rules. I haven't lived in CT since 2011, so this might be out of date. Going west on I-95 from Groton toward NYC, I'd start in a 55 zone, then to a 65 zone, then to a 50 zone, then a 70 zone, then you'd hit New Haven and suddenly I-95 is a 40 MPH zone, before going back to a 55 zone that inexplicably became 50 mph every now and then. (Eventually you'd hit Stamford's asinine traffic and everything might as well be a 5mph zone, but that's a different issue.) The only spot in the whole thing that made sense was the 65-to-50 switchup, because there was a hilariously bad interchange with a right-side entrance and left-side exit within ~0.25 miles of each other. You'd reliably hit slow-moving traffic who just needed to cross the highway to the other exit. This is still more structural than driver issue, though; that sort of on/off design shouldn't exist on an interstate IMO. And yeah, people reliably went 15+ over the limit everywhere unless they saw a cop car. Every state has its own style of driving idiocy, and hilarious speeding was CT's thing.
|
# ? Jun 13, 2019 22:14 |
|
luxury handset posted:humans are generally pretty good drivers so long as they aren't distracted. a huge number of car accidents are from teens, young adult men, old people, and folks who are doing things they shouldn't be doing like eating, speeding, racing, or playing with phone, or intoxicated/impaired. experienced drivers who are attentive and driving at a reasonable speed generally don't get in accidents - this is why insurance companies can hand out discounts for people who go multiple years at a time without getting in accidents (at least ones that are too bad to be ignored or forgiven by the other driver) And technology is really helping with that, too, with features like auto-brake (the thing Uber disabled on their test car that ran over a person), ACC, blind spot monitoring, and all that crap. Instead of relying on technology to do 100%, we can use technology to augment human weaknesses and prevent accidents, while still relying on humans to do the majority of the stuff that would be nearly impossible to program a computer to handle in all situations. Eventually, I see it getting to a point like autopilot in planes, where computers take care of the tricky, monotonous bits that don't require much thinking, and humans can make the decisions and manage the edge cases.
|
# ? Jun 13, 2019 22:23 |
|
PT6A posted:Eventually, I see it getting to a point like autopilot in planes, where computers take care of the tricky, monotonous bits that don't require much thinking, and humans can make the decisions and manage the edge cases. That doesn't make any sense given the allowable reaction times for events on the road versus allowable reaction times for events in the air.
|
# ? Jun 13, 2019 22:32 |
|
shrike82 posted:That doesn't make any sense given the allowable reaction times for events on the road versus allowable reaction times for events in the air. Except computer reaction times are far faster, that's exactly the sort of things that computers can and do already handle in cars. "There's something in front of me, I need to alert the driver or stop urgently if they don't respond" is the sort of thing my car already handles for me in an emergency. The problem with Uber's system is that they disabled that so the car could make more reasoned decisions about whether the thing in front is a threat, whether it will move out of the way in time, etc., and then it hit and killed a woman.
|
# ? Jun 13, 2019 23:00 |
|
PT6A posted:Except computer reaction times are far faster, that's exactly the sort of things that computers can and do already handle in cars. "There's something in front of me, I need to alert the driver or stop urgently if they don't respond" is the sort of thing my car already handles for me in an emergency. The problem with Uber's system is that they disabled that so the car could make more reasoned decisions about whether the thing in front is a threat, whether it will move out of the way in time, etc., and then it hit and killed a woman. I think the objection was more that there won't be many scenarios where, when human input is needed, a person will have time to react given an alert from the system. Split second emergencies requiring human intervention are much more common on the ground than in the air. There Bias Two fucked around with this message at 23:05 on Jun 13, 2019 |
# ? Jun 13, 2019 23:03 |
|
If you're talking about the trivial use-case of augmenting a driver who's always fully in control, what's the point - driving isn't difficult. If you're talking about actual "autopilot" (just see how people are abusing Tesla's autopilot and literally sleeping on the wheel), there's no way of getting around the fact that events on the road can crop up within seconds.
|
# ? Jun 13, 2019 23:07 |
|
Doublepost
|
# ? Jun 13, 2019 23:10 |
|
Hot take: if I commuted by car for 1h or more, I'd accept a cumulative 10-20% risk of death from self driving cars over my working life so I can jerk off instead of having to manually drive the same route forever.
|
# ? Jun 13, 2019 23:10 |
|
shrike82 posted:If you're talking about the trivial use-case of augmenting a driver who's always fully in control, what's the point - driving isn't difficult. Absolutely. But the thing is: computers are already pretty great at split-second intervention, way better than humans are. A car can determine someone's pulled out in front of me and apply the brakes to maximum allowable force much more effectively than I can as a human. They're poor at avoiding situations where split-second intervention is necessary in the first place, which is where the driver being in overall control of the vehicle is still important and useful.
|
# ? Jun 13, 2019 23:12 |
|
PT6A posted:Absolutely. But the thing is: computers are already pretty great at split-second intervention, way better than humans are. A car can determine someone's pulled out in front of me and apply the brakes to maximum allowable force much more effectively than I can as a human. Yeah, I'm going to have to see the receipts on an assertion like this.
|
# ? Jun 13, 2019 23:13 |
|
shrike82 posted:Yeah, I'm going to have to see the receipts on an assertion like this. Radar and/or lidar to auto-braking is a hell of a lot faster than any human. I'm going to guess that tuning the system to avoid false positives is difficult, and you'd also get repeatedly rear-ended from behind when human drivers slam into your emergency-braking rear end.
|
# ? Jun 13, 2019 23:15 |
|
Twerk from Home posted:Radar and/or lidar to auto-braking is a hell of a lot faster than any human. I'm going to guess that tuning the system to avoid false positives is difficult, and you'd also get repeatedly rear-ended from behind when human drivers slam into your emergency-braking rear end. The false positives happen, that's why there's the chime that signals "hey, you have a risk here, consider braking!" comes well before the "holy gently caress you're going to hit something, I'm stopping now!" automatic braking kicks in. It's not perfect, obviously, but anything that can enhance road traffic safety is a step in the right direction. Of course it shouldn't be strictly necessary in the first place, since you should be driving to avoid situations where it's necessary, but we all gently caress up and I don't mind having a system that will save my rear end 80% if I make a mistake, it's better than nothing.
|
# ? Jun 13, 2019 23:19 |
|
Yeah, assertions like that strike me as a layman transferring what humans find easy/difficult to what a model would find easy/difficult. And kinda funny in light of the Boeing 737 Max incidents shrike82 fucked around with this message at 23:21 on Jun 13, 2019 |
# ? Jun 13, 2019 23:19 |
|
Allow me to unveil my latest invention: a camera mounted on the rear-view mirror, recording the driver's face, using machine learning to detect when their attention isn't directed at the road. This will make the car seat literally electroshock the driver's rear end until their attention goes back to the road. In case the valuation planks we just repurpose them as sex toys, win/win
|
# ? Jun 13, 2019 23:31 |
|
Twerk from Home posted:Isn't 20 mph over the limit usually where fines / penalties / points get enormous? 20 over means mandatory jail time in Virginia: https://jalopnik.com/never-speed-in-virginia-lessons-from-my-three-days-in-1613604053 20mph over the limit is normal in So. California.
|
# ? Jun 14, 2019 00:21 |
|
|
# ? Jun 8, 2024 15:27 |
|
VideoGameVet posted:20mph over the limit is normal in So. California. Is there anywhere in the US where it's actually normal to go the posted speed limit?
|
# ? Jun 14, 2019 01:37 |