phantic will be out of a job if tesla is ever able to automate dicksucking. I look forward to the TY for firing me tweet.
|
|
# ? Jun 14, 2018 18:57 |
|
|
# ? Jun 8, 2024 08:18 |
|
Mozi posted:i was taught to not drive over things like a cardboard box in the road because you don't always know if there might be something under it A cardboard box is opaque, so to your eyes there might be something inside it. A balloon is not opaque to ultrasound. A car is. Would you drive through a pile of leaves being blown across the road, or a wave of water being thrown up by a car coming the other way driving through a puddle, or would you slam on your brakes or suddenly swerve? You'd drive through the leaves because you can tell they're not a big solid object that will cause any damage and that just maintaining speed and course is safer than doing literally anything else. That is a decision autopilots will also need to know how to make if they're going to be any good at all; they will need to know the difference between "this thing in the road doesn't warrant taking action for" and "oh poo poo there's something in the road." Concluding that "since the Tesla plowed into a big car-sized balloon it would have done the same thing in a situation where you replaced the balloon with a car" is unsupported by the video. That's my point. It's a lovely test. It's not a meaningful video. It says nothing about a Tesla operating in real-world conditions. This does: https://www.youtube.com/watch?v=-2ml6sjk_8c I am not defending Tesla's autopilot. I am attacking the dumb video with a lovely test.
|
# ? Jun 14, 2018 19:02 |
|
You do realize that all you're proving with this line of argument is that computer vision and sensor integration is an extraordinarily complicated problem that we're nowhere near being able to solve to the degree necessary to put our lives in its hands, right?
|
# ? Jun 14, 2018 19:04 |
|
Phanatic posted:It didn't stop for a balloon, not a car. If they had used a big soap bubble instead of a balloon, would anyone think that test showed anything about anything? No. So why's a balloon any more meaningful a test than a soap bubble? It stopped late for the car, stopping within the envelope of where the car was, it didn't hit it full speed. The Tesla algorithm probably doesn't leave a full stopping distance in front of the car, because no one does that on the road, and Tesla cars would become a nuisance. It does mean that if the car in front evades instead of braking, you'll get an accident. For a car traveling 65 mph, the standard braking distance used in the AASHTO Policy on Geometric Design of Highways, using deceleration of 11.2 ft/s2 (again what AASHTO uses for humans), it would take 405 feet to come to a stop. That means that you'd get about 13 cars per mile, for a throughput of 845 veh/hr/lane, well under the maximum capacity of 1900 veh/hr/lane for human drivers that engineers use now. If you add in some reaction breaking time (say, half a second - we use 2.5s for humans), that goes up to 450 feet of distance, and you're down to 760 veh/hr/lane. Any solution that doesn't rob the roadway of capacity is going to require automated vehicles to be traveling in platoons, communicating with each other, and relying on the car in front to spot stopped vehicles/dangers so that cars behind can take reasonable measures to avoid accidents.
|
# ? Jun 14, 2018 19:06 |
|
haveblue posted:You do realize that all you're proving with this line of argument is that computer vision and sensor integration is an extraordinarily complicated problem that we're nowhere near being able to solve to the degree necessary to put our lives in its hands, right? I'm not sure where you think I've ever argued otherwise. Was it earlier in the thread where people were saying that the LIDAR on the uber car might not have seen the guy on the bike and I was pointing out that, no, LIDAR would definitely been capable of seeing him and therefore something really hosed up was going on with the autopilot and it was really unsafe? In that entire discussion I was pointing out that it's a hard problem, autonomous vehicles aren't even close to being able to handle routine driving conditions, and that I'm really skeptical of them in general, and then I posted the video of a Tesla barreling into a traffic barrier. I'll never not be amazed at the ability of goons to interpret a statement as the literal opposite of what it says. "I'm skeptical of autonomous vehicles," a thing I actually said, turns into "tesla dicksucking." Devor posted:If you add in some reaction breaking time (say, half a second - we use 2.5s for humans), that goes up to 450 feet of distance, and you're down to 760 veh/hr/lane. Thinking you mean .25s here? Phanatic fucked around with this message at 19:17 on Jun 14, 2018 |
# ? Jun 14, 2018 19:13 |
|
i appreciate that the tesla recommendation is that drivers need to be paying attention to the road and ready to intervene in an emergency to avoid a collision if i'm being that vigilant and aware of my surroundings while i'm driving what the gently caress do i need self driving for then
|
# ? Jun 14, 2018 19:14 |
|
Phanatic posted:I'm not sure where you think I've ever argued otherwise. Probably the place where you were arguing that crashing into an object that closely resembled a car in every significant way except density was a poorly designed test and not a failure of the mechanism being tested.
|
# ? Jun 14, 2018 19:15 |
|
Phanatic has an uncanny ability to convince everyone that whatever side of an argument he's arguing for is wrong. Regardless of the topic. And I don't think it's like a deliberate concern trolling thing. But somehow that's not because he's terrible at having a discussion, it's because every other goon gets it wrong.
|
# ? Jun 14, 2018 19:17 |
|
haveblue posted:Probably the place where you were arguing that crashing into an object that closely resembled a car in every significant way except density was a poorly designed test and not a failure of the mechanism being tested. It doesn't closely resemble a car in every significant way. It cosmetically resembles a car. That's not going to closely resemble a car to a whole bunch of sensor technologies, including ultrasound. Which is why it's not a good test of what it purports to be testing.
|
# ? Jun 14, 2018 19:20 |
|
Phanatic posted:It didn't stop for a balloon, not a car. If they had used a big soap bubble instead of a balloon, would anyone think that test showed anything about anything? No. So why's a balloon any more meaningful a test than a soap bubble? lmao
|
# ? Jun 14, 2018 19:24 |
Phanatic posted:It doesn't closely resemble a car in every significant way. It cosmetically resembles a car. That's not going to closely resemble a car to a whole bunch of sensor technologies, including ultrasound. Which is why it's not a good test of what it purports to be testing. so a tesla is by design a murder machine at a parade. good to know
|
|
# ? Jun 14, 2018 19:24 |
|
Phanatic posted:Agreed. The sensors, however, are ultrasound, which can (in principle, I'm not making an assertion about the particular sensors on the Model 3) tell the difference between solid and hollow objects. In this case, one of three things happened: Is a car a solid or a hollow object? What kind of ultrasonic sensors can distinguish hollow and solid objects from a distance? Are you, in fact, talking out of your arse?
|
# ? Jun 14, 2018 19:26 |
|
Submarine Sandpaper posted:so a tesla is by design a murder machine at a parade. good to know elon musk and his engineers on the tesla test track, january 2017 https://www.youtube.com/watch?v=SO9rJtNtmqc&t=74s
|
# ? Jun 14, 2018 19:29 |
Phanatic posted:It doesn't closely resemble a car in every significant way. It cosmetically resembles a car. That's not going to closely resemble a car to a whole bunch of sensor technologies, including ultrasound. Which is why it's not a good test of what it purports to be testing. It doesn't have to? It's an obstacle in the road that the car should be able to react to. Regardless of what it is or what it is made of.
|
|
# ? Jun 14, 2018 19:29 |
|
Phanatic posted:It doesn't closely resemble a car in every significant way. It cosmetically resembles a car. That's not going to closely resemble a car to a whole bunch of sensor technologies, including ultrasound. Which is why it's not a good test of what it purports to be testing. THEN WHY DIDN'T IT MOVE INTO THE EMPTY LANE ON THE RIGHT YOU DOORKNOB
|
# ? Jun 14, 2018 19:33 |
|
Phanatic posted:It doesn't closely resemble a car in every significant way. It cosmetically resembles a car. That's not going to closely resemble a car to a whole bunch of sensor technologies, including ultrasound. Which is why it's not a good test of what it purports to be testing. Then the sensors are inappropriate for this application and this is a failure of Tesla's. There is no way to argue this into Tesla being in the right. The car violated the rules of the road and the principles of basic safety and "its perception is too different from humans to make the same decisions" is a hilariously poor excuse for it. haveblue fucked around with this message at 19:40 on Jun 14, 2018 |
# ? Jun 14, 2018 19:35 |
|
the teslabrain thought it was totally ok to drive straight into a balloon the size of the car itself jeez you guys it would've remained in full control the entire time after that and continued to the destination regardless of having the entire windshield enveloped in mylar it uses ultrasound to see what's the big deal
|
# ? Jun 14, 2018 19:36 |
Though it does bear the question of what technology would be required to keep a car from freaking out and screeching to a halt for a pile of leaves or a plastic bag blowing in the wind.
|
|
# ? Jun 14, 2018 19:39 |
|
Probably radar, like the Model S in the video has used since 2014. Also I'm going to go out on a limb and suggest that maybe the people testing this aren't complete morons and made sure the inflatable balloon car was aluminised, i.e. reflective to radar. GotLag fucked around with this message at 19:46 on Jun 14, 2018 |
# ? Jun 14, 2018 19:41 |
|
Phanatic posted:That's not going to closely resemble a car to a whole bunch of sensor technologies, including ultrasound. Former ultrasonic test technician here popping in to let you know you're 100% wrong about this and apparently have no clue how sound works and based on how wrong you are here you're probably also wrong about everything else in this dumb armchair assessment.
|
# ? Jun 14, 2018 19:54 |
|
chitoryu12 posted:Though it does bear the question of what technology would be required to keep a car from freaking out and screeching to a halt for a pile of leaves or a plastic bag blowing in the wind. the same technology we have now but with another few years of tweaking and refinement tesla and uber are both desperate to reach market first because they might go out of business if they don't get something big, soon the normal car manufacturers as well as waymo aka alphabet aka google are all healthy, functioning businesses so they're not nearly as reckless to push shoddy product
|
# ? Jun 14, 2018 19:58 |
|
FCKGW posted:THEN WHY DIDN'T IT MOVE INTO THE EMPTY LANE ON THE RIGHT YOU DOORKNOB Uh, air is less dense than balloons, so driving into the balloon car was actually the safest course of action.
|
# ? Jun 14, 2018 20:04 |
|
"It detected this object and can tell this is not as solid as a car therefore it is okay to crash into it" is not a good defense of Tesla in the context of that video, imo. On the contrary, I'd be alarmed if that were a judgement they were attempting. I also don't think that's what happened here at all. Don't Teslas also have a camera sensor anyway?
|
# ? Jun 14, 2018 20:15 |
|
Devor posted:For a car traveling 65 mph, the standard braking distance used in the AASHTO Policy on Geometric Design of Highways, using deceleration of 11.2 ft/s2 (again what AASHTO uses for humans), it would take 405 feet to come to a stop. You realize that a car can decelerate at a much higher rate that 11.2 ft/s^2, and an emergency stop doesn't have to follow asshto guidlines, right? Good tires on dry pavement will do around a full g, which gives a stopping distance of more like 100ft excluding the reaction time.
|
# ? Jun 14, 2018 20:19 |
|
Son of Thunderbeast posted:Former ultrasonic test technician here popping in to let you know you're 100% wrong about this and apparently have no clue how sound works and based on how wrong you are here you're probably also wrong about everything else in this dumb armchair assessment. Acoustic waves reflect from impedance mismatches, same as EM waves. When passing from one medium to another, if the characteristic impedance changes from one medium to another, there's a mismatch that results in a reflection, the nature and strength of which is different based on the impedance ratio. Polyurethane and polyethylene rubbers can be made that are so absorbtive to ultrasound that they are used to build anechoic ultrasound materials, or so transparent to ultrasound that they're used as ultrasonic probe covers or windows for ultrasonic emitters. They're also commonly used in balloons. https://www.acoustics.co.uk/product/aptflex-f28/ But a rubber balloon's going to look the same to ultrasound as a steel car body. Gotcha.
|
# ? Jun 14, 2018 20:24 |
|
Boy, that "See you in 10 pages" guy was dead on, huh?
|
# ? Jun 14, 2018 20:29 |
|
What sensors does the Tesla in that test have?
|
# ? Jun 14, 2018 20:30 |
|
Phanatic posted:But a rubber balloon's going to look the same to ultrasound as a steel car body. Gotcha. What do the empty lanes to either side of that balloon look like to ultrasound? How the gently caress do you justify running into *anything* when there is nothing around on either side of the object? Because swerving is too dangerous? Looks like the first car managed it just fine.
|
# ? Jun 14, 2018 20:33 |
|
HardDiskD posted:What sensors does the Tesla in that test have? A forward-looking radar, 12 ultrasonics pointing in all directions in the plane, 3 (I think) cameras pointing forward with varying FOVs, and a rear camera. According to page I'm looking at, the ultrasonics are only for close-range stuff, 8 meters. Radar is for out to 160 meters. Tochiazuma posted:What do the empty lanes to either side of that balloon look like to ultrasound? Very possibly. My supposition here is that the big empty rubber bag showed up on the cameras, but not on the radar or the ultrasound, and a visual return combined with a lack of radar return is interpreted to be not a substantial object. Because it wasn't a substantial object. A big rubber balloon is more substantial than most innocuous road debris, I admit, but I think the assumption that you won't see many big rubber balloons on the road is probably a valid one. quote:Because swerving is too dangerous? More likely because an autopilot that swerves its rider into another lane because it can't tell the difference between a small boulder and an empty grocery bag would not endear itself to its owners. Do you swerve to avoid stuff like that? Phanatic fucked around with this message at 20:44 on Jun 14, 2018 |
# ? Jun 14, 2018 20:38 |
|
boner confessor posted:i appreciate that the tesla recommendation is that drivers need to be paying attention to the road and ready to intervene in an emergency to avoid a collision
|
# ? Jun 14, 2018 20:48 |
|
This derail sucks and gently caress anyone who participated in it.
|
# ? Jun 14, 2018 20:50 |
|
VictorianQueerLit posted:see you guys in ten pages has it been ten pages yet? Probably drat close. gently caress, some of you are insufferable.
|
# ? Jun 14, 2018 20:50 |
|
Phanatic posted:Very possibly. My supposition here is that the big empty rubber bag showed up on the cameras, but not on the radar or the ultrasound, and a visual return combined with a lack of radar return is interpreted to be not a substantial object. Because it wasn't a substantial object. A big rubber balloon is more substantial than most innocuous road debris, I admit, but I think the assumption that you won't see many big rubber balloons on the road is probably a valid one. First, according to this Tesla website (https://www.tesla.com/en_CA/autopilot) the ultrasonics aren't pointing forwards and have a range of 8m so having them not register the object directly in front is not surprising, since that's not what they are there for. Whether or not I swerve to avoid grocery bags, small rocks or any other small object is immaterial since I *would* swerve into a completely open lane to avoid a giant balloon or other large object directly in my path when I drive. You're trying to argue that due to some sort of advanced sensor this car is able to tell it is about to hit a large balloon but not whether the lanes on either side are clear, so that running into a car-sized balloon is somehow better than avoiding it. Which is loving stupid.
|
# ? Jun 14, 2018 20:51 |
|
put rubber baby buggy bumpers on the tesla who the gently caress needs to stop if you can just bounce off
|
# ? Jun 14, 2018 20:51 |
|
I'd go even further and say that if you're not driving you *can't* be as vigilant and aware of your surroundings as if you are driving. Not for any great length of time, anyway. Without the need to constantly be observing and reacting to the dynamic environment you're in, you're eventually going to get lost in your own head. Heck, people do that now even when they are driving, like when they're on a well-known route under standard conditions like a daily commute. "Our autopilot still needs you to pay attention" is a cop-out corporate bullshit excuse for a lovely autopilot. And the "oh poo poo, we're hosed unless you, the driver, do something alarm" happens way too late for the typical driver to do anything with that time interval. Tochiazuma posted:First, according to this Tesla website (https://www.tesla.com/en_CA/autopilot) the ultrasonics aren't pointing forwards and have a range of 8m so having them not register the object directly in front is not surprising, since that's not what they are there for. According to that website they're pointing in all directions; if you look at the color-coded diagram the ultrasound-colored region surrounds the entire car. quote:Whether or not I swerve to avoid grocery bags, small rocks or any other small object is immaterial since I *would* swerve into a completely open lane to avoid a giant balloon or other large object directly in my path when I drive. You're trying to argue that due to some sort of advanced sensor this car is able to tell it is about to hit a large balloon but not whether the lanes on either side are clear, so that running into a car-sized balloon is somehow better than avoiding it. No, I'm arguing that the system can tell the difference between "grocery bag" and "rock the same size as a grocery bag" and that the software-based heuristic that an object giving a visual return but showing no radar return is an obstacle that does not demand a sudden evasive maneuver is a much safer assumption on a real road because on a real road giant balloons are only found very very rarely. I guarantee you can come up with an artificial scenario that violates the understanding that went into developing the software and would flummox any autopilot (or software system in general), but that the system was not designed with that artificial scenario in place does not make it loving stupid. It's *possible* that it's loving stupid. Like, if it were the same thing as the Uber fatality: 1. The Uber car detected an obstacle and even classified it as a bicycle(!) 2. Determined that emergency braking was required 3. Couldn't emergency brake under computer control 4. Had no provision to alert the driver that emergency braking was necessary (also !) *That* would be loving stupid. And it's entirely possible that that's what happened as well, and if Tesla's software worked the same way, yes, loving stupid. But the video doesn't go into any kind of detail, it just shows an edited sequence of the car striking the balloon. Phanatic fucked around with this message at 21:08 on Jun 14, 2018 |
# ? Jun 14, 2018 20:52 |
|
moist turtleneck posted:put rubber baby buggy bumpers on the tesla who the gently caress needs to stop if you can just bounce off https://www.youtube.com/watch?v=LlyEIaRw010
|
# ? Jun 14, 2018 20:54 |
|
if the autopilot has an airframe parachute does that mean it orders its steak well done
|
# ? Jun 14, 2018 20:56 |
|
only if the airframe is circumcised
|
# ? Jun 14, 2018 20:59 |
|
"A jaywalker simply can't be run over...and sometimes, that's more than he deserves." Harsh, but fair.
|
# ? Jun 14, 2018 21:08 |
|
|
# ? Jun 8, 2024 08:18 |
|
Threading title still holding up.
|
# ? Jun 14, 2018 21:09 |