Forum adverts like this one are shown to any user who is not logged in. Join us by filling out a tiny 3 field form and you will get your own, free, dakka user account which gives a good range of benefits to you:
No adverts like this in the forums anymore.
Times and dates in your local timezone.
Full tracking of what you have read so you can skip to your first unread post, easily see what has changed since you last logged in, and easily see what is new at a glance.
Email notifications for threads you want to watch closely.
Being a part of the oldest wargaming community on the net.
If you are already a member then feel free to login now.
What's extra hilarious about driverless cars is that we have already invented public transportation. Cutting down on the amount of cars out there is good but we've already got a more efficient method of doing so than replacing them 1:1 with Google knock-off robots.
Grey Templar wrote: So if something unexpected happens, like say a sudden lane change from another vehicle, a human might be able to compensate while a driverless car might not compensate enough.
It's actually the exact opposite. This situation overwhelmingly favors the automated car, not the human. A sudden lane change is unpredictable, but it's still an event that falls well within the set of possible events known to occur on roads, and involves an object that is a clearly visible target whose motion can be very well observed and reacted to. The automated car can immediately recognize the car's motion, analyze exactly what separation distance exists as it changes lanes, and react with the minimum response required to keep separation distances at an appropriate level. The human driver, on the other hand, is likely to be slower to react as their attention can't be everywhere at once (and that's just about where your eyes can be looking, not even considering driving while distracted with passengers/phones/etc), is certainly less able to judge the exact position and velocity of the other car, and is likely to overreact when they are startled by sudden awareness of a car halfway into their lane.
The hard part for automated cars is not other cars, it's the environment. It's things like the situation in the OP, where the target is difficult for the car's vision system to process (a dark and relatively small figure on a dark background) and must be handled with very short separation distances that give virtually nonexistent margins for error. The problem is no longer reacting to another car's motion, it's having to predict what a pedestrian is going to do before they do it because by the time you see them start to move it's often too late to hit the brakes and avoid a collision.
There is no such thing as a hobby without politics. "Leave politics at the door" is itself a political statement, an endorsement of the status quo and an attempt to silence dissenting voices.
Rosebuddy wrote: What's extra hilarious about driverless cars is that we have already invented public transportation. Cutting down on the amount of cars out there is good but we've already got a more efficient method of doing so than replacing them 1:1 with Google knock-off robots.
Rosebuddy wrote: What's extra hilarious about driverless cars is that we have already invented public transportation. Cutting down on the amount of cars out there is good but we've already got a more efficient method of doing so than replacing them 1:1 with Google knock-off robots.
Have you been to the US?
Just because public transit sucks now doesn't mean it has to suck in the future.
We were once so close to heaven, St. Peter came out and gave us medals; declaring us "The nicest of the damned".
“Anti-intellectualism has been a constant thread winding its way through our political and cultural life, nurtured by the false notion that democracy means that 'my ignorance is just as good as your knowledge.'”
feeder wrote: Just because public transit sucks now doesn't mean it has to suck in the future.
It does when the reasons it sucks are largely things that can't be changed. Public transit in the US is difficult because we're so widely dispersed, the number of routes required to serve everyone would be massive and many of those routes would have very few users. Providing funding and respect for public transit doesn't make people magically consolidate into dense urban areas.
There is no such thing as a hobby without politics. "Leave politics at the door" is itself a political statement, an endorsement of the status quo and an attempt to silence dissenting voices.
feeder wrote: Just because public transit sucks now doesn't mean it has to suck in the future.
It does when the reasons it sucks are largely things that can't be changed. Public transit in the US is difficult because we're so widely dispersed, the number of routes required to serve everyone would be massive and many of those routes would have very few users. Providing funding and respect for public transit doesn't make people magically consolidate into dense urban areas.
Sure, but 90% of traffic problems occur in those dense urban areas, which would be alleviated by well-funded quality public transport.
We were once so close to heaven, St. Peter came out and gave us medals; declaring us "The nicest of the damned".
“Anti-intellectualism has been a constant thread winding its way through our political and cultural life, nurtured by the false notion that democracy means that 'my ignorance is just as good as your knowledge.'”
feeder wrote: Just because public transit sucks now doesn't mean it has to suck in the future.
It does when the reasons it sucks are largely things that can't be changed. Public transit in the US is difficult because we're so widely dispersed, the number of routes required to serve everyone would be massive and many of those routes would have very few users. Providing funding and respect for public transit doesn't make people magically consolidate into dense urban areas.
Sure, but 90% of traffic problems occur in those dense urban areas, which would be alleviated by well-funded quality public transport.
Exactly. An autonomous car will just be a very high-tech set of wheels in the same traffic jam as the other ones.
Autonomous ride-sharing vehicles make sense in low to medium density areas. Go a bit further and vehicle footprint is an issue and then you're talking buses and trams.
schadenfreude wrote: Driverless cars will always kill people no matter how good the technology become. The low hurdle it must overcome is killing less people than human drivers. That's not going to be difficult when you consider the number of glitches humans have such as drunk driving, txting while driving, falling asleep while driving, road rage, and street racing. Driving will easily be safer once our robotic overlords control our vehicles.
There is a caveat to that. Only if the vast majority/all of the other cars on the road are also driverless cars and all of them communicate with each other.
Actually the low bar I was talking about was the hybrid environment of driver and driverless cars that you just described. Even in that worse case scenario where humans are causing problems the AI isn't drunk driving, txting while driving, falling asleep while driving, road rage, and street racing. The AI will initially take a lot of blame but they are backed by powerful multinational corporations who also have the facts on their side. Overall traffic deaths wI'll go down in that worst case scenario you are imagining because human being are terrible drivers.
There is also some Orwellian angles we are not looking at. Once the AI start to talk to each other one of the first things they will be programed to do is detect patterns in human drivers that indicate that they are drunk, avoid the drunk driver giving them a wide distance, warn other AI of the drunk driver, and the AI is going to snitch. Pretty soon every self driving vehicle is going to be a highway patrol observation drone for drunk drivers, and nobody will care because drunk drivers are scum that endanger the lives of everyone we care about.
Chaos isn’t a pit. Chaos is a ladder. Many who try to climb it fail, and never get to try again. The fall breaks them. And some are given a chance to climb, but refuse. They cling to the realm, or love, or the gods…illusions. Only the ladder is real. The climb is all there is, but they’ll never know this. Not until it’s too late.
So about this case specifically, they released the dash-cam video, and I can say quite confidently that had I been driving, I would not have been able to stop in time from when she is first spotted. Now that doesn't mean the AI shouldn't have been able to, they should have been able to install some sort of sensor to allow the car to see what headlights don't illuminate, but she just appears so suddenly, no human driver would have been able to stop in time. my $0.02 anyway
Homosexuality is the #1 cause of gay marriage.
kronk wrote: Every pizza is a personal sized pizza if you try hard enough and believe in yourself.
sebster wrote: Yes, indeed. What a terrible piece of cultural imperialism it is for me to say that a country shouldn't murder its own citizens
BaronIveagh wrote: Basically they went from a carrot and stick to a smaller carrot and flanged mace.
Wait, does a robot need light to 'see'? I thought they used radar.
We were once so close to heaven, St. Peter came out and gave us medals; declaring us "The nicest of the damned".
“Anti-intellectualism has been a constant thread winding its way through our political and cultural life, nurtured by the false notion that democracy means that 'my ignorance is just as good as your knowledge.'”
BobtheInquisitor wrote: Would a human driver have slowed down or turned on high beams in anticipation of low light visibility issues? My experience says, "yes."
Out of interest, on a road without crossings how much slower than the speed limit do you drive at night? Espically when it was following behind another car at a safe distance. And there was a car in front, so putting high-beams on is pretty discouraged (generally it's assumed that the car in front will have high beams on and will stop/slow down in the case of an obsticle, an high beams from behind just blind drivers in front). You can sort of see her feet get illuminated at 18 seconds in the video, but wearing all black she doesn't really come into view until 19 and is hit on 20. The issue, more than anything, is that she shouldn't have tried to run in-between two cars, at night, while wearing all black. Now a (non-distracted) human driver would have slammed on the breaks earlier, but she would have still gotten hit.
Automatically Appended Next Post:
feeder wrote: Wait, does a robot need light to 'see'? I thought they used radar.
Apparently not anything good enough. Robots *should* be better than humans. If the AI had the same reflexes as a human but no issues with darkness this accident could have been easily avoided. And I have seen stuff showing their ability to stop in day that would have resulted in no deaths. But apparently they skimped on giving it proper night vision or something.
This message was edited 1 time. Last update was at 2018/03/22 22:23:37
Homosexuality is the #1 cause of gay marriage.
kronk wrote: Every pizza is a personal sized pizza if you try hard enough and believe in yourself.
sebster wrote: Yes, indeed. What a terrible piece of cultural imperialism it is for me to say that a country shouldn't murder its own citizens
BaronIveagh wrote: Basically they went from a carrot and stick to a smaller carrot and flanged mace.
BobtheInquisitor wrote: Would a human driver have slowed down or turned on high beams in anticipation of low light visibility issues? My experience says, "yes."
Out of interest, on a road without crossings how much slower than the speed limit do you drive at night? Espically when it was following behind another car at a safe distance. And there was a car in front, so putting high-beams on is pretty discouraged (generally it's assumed that the car in front will have high beams on and will stop/slow down in the case of an obsticle, an high beams from behind just blind drivers in front). You can sort of see her feet get illuminated at 18 seconds in the video, but wearing all black she doesn't really come into view until 19 and is hit on 20. The issue, more than anything, is that she shouldn't have tried to run in-between two cars, at night, while wearing all black. Now a (non-distracted) human driver would have slammed on the breaks earlier, but she would have still gotten hit.
Automatically Appended Next Post:
feeder wrote: Wait, does a robot need light to 'see'? I thought they used radar.
Apparently not anything good enough. Robots *should* be better than humans. If the AI had the same reflexes as a human but no issues with darkness this accident could have been easily avoided. And I have seen stuff showing their ability to stop in day that would have resulted in no deaths. But apparently they skimped on giving it proper night vision or something.
I admit I haven't seen the video. I don't want to watch someone run down. If there was other traffic, and she was trying to run between cars, then that is new information for me. before, it sounded like she was trying to cross in a dark stretch of road, not dart across a busy stretch.
And for the record, I slow down quite a bit on dark country roads.
feeder wrote:Wait, does a robot need light to 'see'? I thought they used radar.
Cameras, lidar (and other types of lasers), any other sensor that their specific approach needs. There was recently a breakthrough for lidar that made better miniaturisation (and cost saving) possible so we'll get cars with better and better assisted driving features while not even seeing where the sensors are (not like test vehicles that have camera/sensor arrays on top). I think right now Tesla relies on a lot of regular cameras for their "self driving" feature which tends to have problems with snow. If I remember correctly the Tesla with the fatal crash happened because the car couldn't "see" the side of a big white truck trailer. The driver was too optimistic about the "assisted driving feature" and was watching a movie while his car saw the big trailer as "open road" and just drove on/into it.
feeder wrote: Wait, does a robot need light to 'see'? I thought they used radar.
My car is not autonomous, but it uses both radar and optical. There is a camera that detects cars in front of it for the purposes of keeping a set distance when cruise control is on, and it also will brake automatically if it detects a crash is imminent and the brakes are not already being applied. Radar mostly is used to monitor the blind spots on either side.
lord_blackfang wrote: Respect to the guy who subscribed just to post a massive ASCII dong in the chat and immediately get banned.
Flinty wrote: The benefit of slate is that its.actually a.rock with rock like properties. The downside is that it's a rock
BobtheInquisitor wrote: Would a human driver have slowed down or turned on high beams in anticipation of low light visibility issues? My experience says, "yes."
Out of interest, on a road without crossings how much slower than the speed limit do you drive at night? Espically when it was following behind another car at a safe distance. And there was a car in front, so putting high-beams on is pretty discouraged (generally it's assumed that the car in front will have high beams on and will stop/slow down in the case of an obsticle, an high beams from behind just blind drivers in front). You can sort of see her feet get illuminated at 18 seconds in the video, but wearing all black she doesn't really come into view until 19 and is hit on 20. The issue, more than anything, is that she shouldn't have tried to run in-between two cars, at night, while wearing all black. Now a (non-distracted) human driver would have slammed on the breaks earlier, but she would have still gotten hit.
Automatically Appended Next Post:
feeder wrote: Wait, does a robot need light to 'see'? I thought they used radar.
Apparently not anything good enough. Robots *should* be better than humans. If the AI had the same reflexes as a human but no issues with darkness this accident could have been easily avoided. And I have seen stuff showing their ability to stop in day that would have resulted in no deaths. But apparently they skimped on giving it proper night vision or something.
I admit I haven't seen the video. I don't want to watch someone run down. If there was other traffic, and she was trying to run between cars, then that is new information for me. before, it sounded like she was trying to cross in a dark stretch of road, not dart across a busy stretch.
And for the record, I slow down quite a bit on dark country roads.
Not super busy, the other car was a bit ahead (not highway close but close enough that if you see one car you see the other, and it would be rude for the AI car to put highbeams on), but it was a very dumb time to run across. If she had waited she would still be alive
Homosexuality is the #1 cause of gay marriage.
kronk wrote: Every pizza is a personal sized pizza if you try hard enough and believe in yourself.
sebster wrote: Yes, indeed. What a terrible piece of cultural imperialism it is for me to say that a country shouldn't murder its own citizens
BaronIveagh wrote: Basically they went from a carrot and stick to a smaller carrot and flanged mace.
I just watched it. There is no chance a human driver could have prevented that. There are less than two seconds between the instant you can see that something is on the road (you see her feet first and then for some time basically nothing else) and the impact. Even if the safety driver had been totally alert, I don't see her reacting fast enough to even slow down the vehicle notably. And tbh, a human watching over a computer who has to interpret if the computer is doing everything right will have a crappy reaction time because he will have to first consciously understand that the vehicle is doing something wrong and then take over. That takes ages compared to a person driving themselves hitting the brakes because there is no "intuition", i.e. unconscious predictions etc. at work there, but actual thinking.
I think, however, that a vehicle with radar and/or infrared should have seen her. There's very deep shadow, but no physical obstruction on the road that would block the signal.
This message was edited 1 time. Last update was at 2018/03/23 01:22:47
I think we'll see a global conflict that culls 1/3 or more of the world population before we see fan wielding robots feed us bon bons while we make the 1% run on hamster wheels.
The rich would indeed rather kill a third of the planet than share wealth.
I'm shocked they don't physically come into your house and take your stuff so they can be richer.
And when that stuff goes down, bombs don't discriminate between the evil rich people that are subjugating the populace (in your mind) and the general populace. Half the fun will be trying to rebuild the countries involved, and I'm fairly confident that some of the state actors in play right now are getting brave enough to attempt a larger scale attack on countries that it would have been unthinkable to attack decades ago.
I didn't want to watch it, because it seems macabre, but I thought it important to see the evidence of the circumstances of the collision for myself.
Having seen it, I think that part of the blame must lie with the victim. She was pushing her bike across a dark, 40 mph street, with no lights or hi-vis clothing.
I'm not surprised the car did not stop by itself if it relies on radar. A bike is a pretty thin, small object on radar, I imagine. I am a rather more surprised that the safety driver did not see the cyclist and stop the car. However, she seems not to have been paying attention to the road.
The road was dark, the headlights seemed to be in low beam, and the cyclist was wearing dark clothing.
I'm not sure I could have stopped even if I was driving car myself assuming I didn't know in advance there was going to be such a case(ie if I would now enter simulated test to see if I could stop I would have pre-knowledge I'll be needing to break quickly which is going to increase my chances dramatically. On real life this would be total surprise). Would be touch&go unless that video is sped up from real speed. Seems it was about 1 sec from where I got first hint and where the hit happened. Wasn't human reaction time about 1 sec?
This message was edited 1 time. Last update was at 2018/03/23 06:50:07
Perhaps adding a FLIR to pick up body heat? The range isn't always the best on those, but any little bit would help. Either that, or we force people to add plating to bikes to widen the profile so radar can pick it up easier.
XuQishi wrote: I just watched it. There is no chance a human driver could have prevented that. There are less than two seconds between the instant you can see that something is on the road (you see her feet first and then for some time basically nothing else) and the impact. Even if the safety driver had been totally alert, I don't see her reacting fast enough to even slow down the vehicle notably. And tbh, a human watching over a computer who has to interpret if the computer is doing everything right will have a crappy reaction time because he will have to first consciously understand that the vehicle is doing something wrong and then take over. That takes ages compared to a person driving themselves hitting the brakes because there is no "intuition", i.e. unconscious predictions etc. at work there, but actual thinking.
Auto braking did help this Norwegian girl.
Uber cars have both radar and lidar (which don't care about light conditions), and the situation is an ideal test scenario for them, an object slowly crossing the road at right angles, right in the center, with no other interfering objects in a very simple environment.
Either the sensors or the software responsible for interpreting them failed big time.
I know it sounds harsh, but these are, ultimately, just teething troubles.
Now there's a strong argument something like this should've been identified and sorted prior to on-the-road testing/deployment. But there's no argument whatsoever that it's a sign the technology has an insurmountable, unavoidable flaw.
Fed up of Scalpers? But still want your Exclusives? Why not join us?
Kilkrazy wrote: A human driver who was paying attention would have had two seconds to spot the cyclist and react.
Key point: paying attention. This is a bad assumption to make.
There is no such thing as a hobby without politics. "Leave politics at the door" is itself a political statement, an endorsement of the status quo and an attempt to silence dissenting voices.
I really dislike the idea of self-driving cars, but I can't blame the AI here. I would have totally crashed into that woman as well. She appears way too suddenly to stop in time. And it seems to be a large, straight road where you'd not expect people to suddenly cross. No reason to limit your speed.
That auto-break system Jouso's video shows should be mandatory though. It just stops a heavy, speeding truck in like 2 seconds! That is amazing! It saved that kid (though he must have gotten the scare of his life...) and it could probably safe a lot more lives if it was implemented in all cars.
Mad Doc Grotsnik wrote: I know it sounds harsh, but these are, ultimately, just teething troubles.
Now there's a strong argument something like this should've been identified and sorted prior to on-the-road testing/deployment. But there's no argument whatsoever that it's a sign the technology has an insurmountable, unavoidable flaw.
Or is it? That assumes human driver would succeed where ai failed. I still would bet in favour of ai succeeding more often than human
The key advantage of the AI is its ability to react faster than a person. When the detection equipment is on par or superior to a human then the machine can react far quicker than a human. Furthermore the machine won't in any way panic. As a result things like sudden deceleration should be done "properly".
A person will not only have potentially a greater lag time on reaction, but can also panic and might well do the wrong thing; or do the correct action but in the wrong way. This is especially true if the person has no prior experience of such an event happening to them before.
The key element with the machine is its ability to see and interpret the information it can see. Indeed its often in this interpretation area where the problems lay (eg the afore mentioned situation where the machine saw the white lorry and interpreted that information incorrectly and drove into it thinking it was clear and open roadway).
