Coyote V. Tesla
March 17, 2025 7:25 AM Subscribe
Former NASA engineer, squirrel competition organizer, and porch pirate antagonist Mark Rober demonstrates the inadequacy of Tesla's driving automation by means of a test that is quite literally Looney Tunes.
The test was to show why Tesla's vision-only strategy for detecting obstacles is inadequate for the job, and why sensors like LIDAR are needed. To do so, he set up a number of tests, culminating in a breakaway wall painted with an image of the road, classic cartoon style, to see if the car would detect if something was off.
The test was to show why Tesla's vision-only strategy for detecting obstacles is inadequate for the job, and why sensors like LIDAR are needed. To do so, he set up a number of tests, culminating in a breakaway wall painted with an image of the road, classic cartoon style, to see if the car would detect if something was off.
I'd forgotten about Musk's resistance to LIDAR until reading about this over the weekend. It's an odd thing--why wouldn't you want your machines of the future to have a vast and interesting sensorium? Considering the influence The Culture had on him, it's doubly odd; you can't imagine a Culture ship that only has eyes and ears.
posted by mittens at 7:43 AM on March 17 [9 favorites]
posted by mittens at 7:43 AM on March 17 [9 favorites]
Camera only is cheaper.
posted by subdee at 7:49 AM on March 17 [49 favorites]
posted by subdee at 7:49 AM on March 17 [49 favorites]
This would be hilarious if not for the implications it has for other vehicles and pedestrians/cyclists.
posted by tommasz at 7:50 AM on March 17 [6 favorites]
posted by tommasz at 7:50 AM on March 17 [6 favorites]
Dont bet on self-driving cars. There's gonna be so many unemployed shortly that there will be no lack of human drivers available, at a much lower cost. And a self-driving car won't carry your groceries in. /s
[yeah, we're still pretty far away from self-driving cars taking over.]
posted by Artful Codger at 7:54 AM on March 17 [4 favorites]
[yeah, we're still pretty far away from self-driving cars taking over.]
posted by Artful Codger at 7:54 AM on March 17 [4 favorites]
Golly, I wonder why this is only coming out now. It's not like they haven't been using a problematic vision-only solution for years.
I think this video says much more about the value of the brand than it does about the technology. If Tesla's reputation has become so radioactive that science-bro youtubers are comically smashing them into brick walls instead of stanning for them, then Tesla's in some serious trouble.
This video could have been made much earlier, but I doubt someone like Mark Rober would've had the financial incentive to do it.
posted by RonButNotStupid at 7:57 AM on March 17 [28 favorites]
I think this video says much more about the value of the brand than it does about the technology. If Tesla's reputation has become so radioactive that science-bro youtubers are comically smashing them into brick walls instead of stanning for them, then Tesla's in some serious trouble.
This video could have been made much earlier, but I doubt someone like Mark Rober would've had the financial incentive to do it.
posted by RonButNotStupid at 7:57 AM on March 17 [28 favorites]
dammit I was thinking of posting this....! Glad it got posted though!
posted by EmpressCallipygos at 8:09 AM on March 17 [2 favorites]
posted by EmpressCallipygos at 8:09 AM on March 17 [2 favorites]
There are plenty of bad drivers on the road, and a good chunk of humanity would probably drive into a matte painting, possibly including myself. Scoring points on Elon Musk is fun and all, but this xkcd seems prescient.
posted by ockmockbock at 8:15 AM on March 17 [6 favorites]
posted by ockmockbock at 8:15 AM on March 17 [6 favorites]
I like that they pre-cut through the foam wall backing the giant picture of a road, so that there was a huge jagged cartoon hole instead of the whole thing buckling and fragmenting. Very nice touch.
posted by egypturnash at 8:22 AM on March 17 [27 favorites]
posted by egypturnash at 8:22 AM on March 17 [27 favorites]
There are plenty of bad drivers on the road, and a good chunk of humanity would probably drive into a matte painting…The problem is not that Teslas drive into matte paintings, but that the same fatal design flaw is why Teslas drive into poles and semis.
posted by ArmandoAkimbo at 8:30 AM on March 17 [23 favorites]
My hope is that all the laughter and mockery will cause Musk's head to asplode in a manner reminiscent of a cartoon character.
posted by BlueHorse at 8:38 AM on March 17 [9 favorites]
posted by BlueHorse at 8:38 AM on March 17 [9 favorites]
There used to be a parking lot in Portland that had one of these painted on the back wall of the lot. 😶
posted by funkaspuck at 8:42 AM on March 17 [3 favorites]
posted by funkaspuck at 8:42 AM on March 17 [3 favorites]
Fun video! Was interesting enough to watch the whole thing, despite containing Disney-related content and that brodude being kind of annoying.
But yeah, Teslas, right? Unbelievable, if not downright criminally negligent, that anything "self driving" does not have lidar. I passed someone fully engrossed in their phone in their Model 3 or whatever. I'd hoped they were at least using autopilot, but even that doesn't inspire very much confidence.
posted by slogger at 8:45 AM on March 17 [2 favorites]
But yeah, Teslas, right? Unbelievable, if not downright criminally negligent, that anything "self driving" does not have lidar. I passed someone fully engrossed in their phone in their Model 3 or whatever. I'd hoped they were at least using autopilot, but even that doesn't inspire very much confidence.
posted by slogger at 8:45 AM on March 17 [2 favorites]
Maybe lidar would have caught this, but its limitations on refresh rate, fidelity at distance, and atmospheric quality are pretty profound, so I'd have liked a demo of that actually working in practise. Not that I'm defending Tesla, just that I don't think anyone's approach to AI driving is particularly good, and I don't think sensors are the biggest issue. FWIW, I do think that including some simpler low-level image processing in the stack (Brooks suggests looming here) might work pretty well for the ACME problem presented, but in general I think the (deadly) edge cases for automated driving are probably AI-Hard.
posted by rhamphorhynchus at 8:47 AM on March 17 [3 favorites]
posted by rhamphorhynchus at 8:47 AM on March 17 [3 favorites]
There are plenty of bad drivers on the road, and a good chunk of humanity would probably drive into a matte painting,
Counterpoint: MassDOT will grind the surface off a multi-lane highway and leave the roadway with only the faintest of fossilized lines to guide anyone. And despite this, the percentage of human drivers who can still successfully navigate is so high that that they'll happily leave it like that for days until it's time to begin paving.
posted by RonButNotStupid at 8:48 AM on March 17 [19 favorites]
Counterpoint: MassDOT will grind the surface off a multi-lane highway and leave the roadway with only the faintest of fossilized lines to guide anyone. And despite this, the percentage of human drivers who can still successfully navigate is so high that that they'll happily leave it like that for days until it's time to begin paving.
posted by RonButNotStupid at 8:48 AM on March 17 [19 favorites]
Musk don't care, Musk don't give a shit...
(Apologies to the Honey Badgers in our audience)
Tesla seems pretty fucked, stock wise and moving forward. We own two, but will not buy another. That charging network tho...
And I can't see how any of the right-wingers who are all in on the DOGE shit will pick up the slack. EVs are for woke liberals. More oil drilling!
Which seems opposite from what Tesla was doing. Ketamine is a hellofa drug I guess.
Nice to see how much cash Elno has lost though. We should keep doing that.
posted by Windopaene at 8:50 AM on March 17 [9 favorites]
(Apologies to the Honey Badgers in our audience)
Tesla seems pretty fucked, stock wise and moving forward. We own two, but will not buy another. That charging network tho...
And I can't see how any of the right-wingers who are all in on the DOGE shit will pick up the slack. EVs are for woke liberals. More oil drilling!
Which seems opposite from what Tesla was doing. Ketamine is a hellofa drug I guess.
Nice to see how much cash Elno has lost though. We should keep doing that.
posted by Windopaene at 8:50 AM on March 17 [9 favorites]
I believe that enough human drivers suck (either because they are drunk or high, reckless, incompetent, or distracted by their phones) that, especially with continued improvements, self-driving cars will be safer than human drivers, either now or in the near future.
But "better-than-human on average" is a low bar. The target should not be "let's get better than where we are," it should be to minimize the number of people maimed or killed by making the new cars as safe as possible.
posted by Mr.Know-it-some at 8:56 AM on March 17 [1 favorite]
But "better-than-human on average" is a low bar. The target should not be "let's get better than where we are," it should be to minimize the number of people maimed or killed by making the new cars as safe as possible.
posted by Mr.Know-it-some at 8:56 AM on March 17 [1 favorite]
Maybe lidar would have caught this, but its limitations on refresh rate, fidelity at distance, and atmospheric quality are pretty profound, so I'd have liked a demo of that actually working in practise.
Maybe I'm misreading what you're saying, but the video explicitly includes a Tesla vs. Lidar shoot-out in a variety of scenarios and lidar caught the cartoon wall just fine.
posted by BlueDuke at 8:59 AM on March 17 [35 favorites]
Maybe I'm misreading what you're saying, but the video explicitly includes a Tesla vs. Lidar shoot-out in a variety of scenarios and lidar caught the cartoon wall just fine.
posted by BlueDuke at 8:59 AM on March 17 [35 favorites]
Considering the influence The Culture had on him
Bah, I will bet dollars to donuts that Musk has never read a Culture novel (or, perhaps, any novel), someone just told him, "Oh you'd like it, the society is ruled by all-powerful AI and the ships have really funny names like "Just Read The Instructions"
posted by Saxon Kane at 9:07 AM on March 17 [28 favorites]
Bah, I will bet dollars to donuts that Musk has never read a Culture novel (or, perhaps, any novel), someone just told him, "Oh you'd like it, the society is ruled by all-powerful AI and the ships have really funny names like "Just Read The Instructions"
posted by Saxon Kane at 9:07 AM on March 17 [28 favorites]
I seem to remember there were possible interference issues with lidar if multiple are operating the same vicinity. Maybe they're not a solution either, if the cameras aren't enough and lidar isn't it, well lets wait until we have a real solution we can wait.
The biggest problem to me remains the semi-fraudulent marketing and the lack of serious oversight.
posted by WaterAndPixels at 9:10 AM on March 17 [4 favorites]
The biggest problem to me remains the semi-fraudulent marketing and the lack of serious oversight.
posted by WaterAndPixels at 9:10 AM on March 17 [4 favorites]
yeah, Musk is no Culture follower, he's a Randian. He thinks he's John Galt.
self-driving cars will be safer than human drivers, either now or in the near future.
No one can claim that yet, until self-driving cars can successfully navigate 99.9% of the road situations with a better score than humans currently do. At present self-driving cars are only safe enough for a restricted subset of roads and driving conditions.
posted by Artful Codger at 9:17 AM on March 17 [7 favorites]
self-driving cars will be safer than human drivers, either now or in the near future.
No one can claim that yet, until self-driving cars can successfully navigate 99.9% of the road situations with a better score than humans currently do. At present self-driving cars are only safe enough for a restricted subset of roads and driving conditions.
posted by Artful Codger at 9:17 AM on March 17 [7 favorites]
That charging network tho...
A consortium of basically everyone-but-Tesla* is building out a competitive network called Ionna.
* BMW, General Motors, Honda, Hyundai, Kia, Mercedes-Benz, Stellantis, and Toyota
posted by jedicus at 9:18 AM on March 17 [10 favorites]
A consortium of basically everyone-but-Tesla* is building out a competitive network called Ionna.
* BMW, General Motors, Honda, Hyundai, Kia, Mercedes-Benz, Stellantis, and Toyota
posted by jedicus at 9:18 AM on March 17 [10 favorites]
Re: Musk and the Culture...
He's Otto from A Fish Called Wanda.
He read some Culture novels and his reading comprehension is akin to Otto determining that the central thesis of Buddhism is "Every man for himself."
posted by ursus_comiter at 9:21 AM on March 17 [30 favorites]
He's Otto from A Fish Called Wanda.
He read some Culture novels and his reading comprehension is akin to Otto determining that the central thesis of Buddhism is "Every man for himself."
posted by ursus_comiter at 9:21 AM on March 17 [30 favorites]
It's the SuperChargers that make Teslas a viable long-distance charging solution. 30 minutes to get filled up? That's doable.
6 hours? Not so much.
Can't tell from that Ionna link what power they are using. How fast the charging will be.
posted by Windopaene at 9:23 AM on March 17
6 hours? Not so much.
Can't tell from that Ionna link what power they are using. How fast the charging will be.
posted by Windopaene at 9:23 AM on March 17
This isn't about self-driving cars though, it's about the Autonomous Emergency Braking that is mandatory in the biggest single car market. (And which I did get beeping at me in the Stellantis/Netherleyland/Vauxhall car I was driving on Saturday — it was a traffic queue and I was just about to go on the brakes, though)
posted by ambrosen at 9:23 AM on March 17
posted by ambrosen at 9:23 AM on March 17
I think he's using Autopilot and not FSD and I don't know why.
I'm not sure what these tests are actually measuring. It's not just LiDAR (actually LiDAR + camera) vs. camera, but the whole tech stack comes into play. Are the problems an issue with "camera only systems" or is the issue that the Tesla's decision making sucks (both are bad, but bad in different ways)?
Put more plainly, it's possible that both the Tesla and whatever-that-other-car-is are identifying objects correctly, but the Tesla is making incredibly bad choices about what to do about it.
One interesting detail about Tesla's camera only system is that it doesn't use parallax to determine distance. It uses the identification of the object to determine a distance (that looks like car, but it's tiny so it must be far away). They used to have multiple cameras (short, medium, and long range) but got rid of them. That might prove to be the bigger error.
I also think it's relevant to compare with a human driver. Even if LiDAR were considerably better than Vision Only (or the particular variation of Vision Only that Tesla is using), if both are light years ahead of humans then it hardly matters. Sure, Stockfish plays better chess than Hedgehog (to pull a name at random), but both of them can kick the living crap out of 99% of humans, and if the worse solution is cheaper and thus can be installed much more easily, what's the issue?
posted by It's Never Lurgi at 9:24 AM on March 17 [3 favorites]
I'm not sure what these tests are actually measuring. It's not just LiDAR (actually LiDAR + camera) vs. camera, but the whole tech stack comes into play. Are the problems an issue with "camera only systems" or is the issue that the Tesla's decision making sucks (both are bad, but bad in different ways)?
Put more plainly, it's possible that both the Tesla and whatever-that-other-car-is are identifying objects correctly, but the Tesla is making incredibly bad choices about what to do about it.
One interesting detail about Tesla's camera only system is that it doesn't use parallax to determine distance. It uses the identification of the object to determine a distance (that looks like car, but it's tiny so it must be far away). They used to have multiple cameras (short, medium, and long range) but got rid of them. That might prove to be the bigger error.
I also think it's relevant to compare with a human driver. Even if LiDAR were considerably better than Vision Only (or the particular variation of Vision Only that Tesla is using), if both are light years ahead of humans then it hardly matters. Sure, Stockfish plays better chess than Hedgehog (to pull a name at random), but both of them can kick the living crap out of 99% of humans, and if the worse solution is cheaper and thus can be installed much more easily, what's the issue?
posted by It's Never Lurgi at 9:24 AM on March 17 [3 favorites]
He's Otto from A Fish Called Wanda.
Otto: *Apes* don't read philosophy.
Wanda: Yes they do, Otto. They just don't understand it!
I think that applies to the muskrat pretty well.
posted by Ickster at 9:28 AM on March 17 [7 favorites]
Otto: *Apes* don't read philosophy.
Wanda: Yes they do, Otto. They just don't understand it!
I think that applies to the muskrat pretty well.
posted by Ickster at 9:28 AM on March 17 [7 favorites]
Can't tell from that Ionna link what power they are using. How fast the charging will be.
"Ionna's DC fast-chargers, which the company calls Ionna Genuine Charge Dispensers, operate at up to 400 kW or up to 200 kW when charging two vehicles at once." Somewhere between comparable-to or faster-than a Supercharger, depending on the Supercharger version.
posted by jedicus at 9:30 AM on March 17 [3 favorites]
"Ionna's DC fast-chargers, which the company calls Ionna Genuine Charge Dispensers, operate at up to 400 kW or up to 200 kW when charging two vehicles at once." Somewhere between comparable-to or faster-than a Supercharger, depending on the Supercharger version.
posted by jedicus at 9:30 AM on March 17 [3 favorites]
I mean, it's not complicated: cars on the EU and UK markets are obliged to have Autonomous Emergency Braking and they should be good enough. But also, the Model Y's works fine.
posted by ambrosen at 9:33 AM on March 17
posted by ambrosen at 9:33 AM on March 17
Maybe I'm misreading what you're saying
Nope, fair enough, I just skimmed the video and saw a load of actual Model 3 and theoretical Lidar. My mistake.
Lidar still doesn't work in the rain though.
posted by rhamphorhynchus at 9:35 AM on March 17
Nope, fair enough, I just skimmed the video and saw a load of actual Model 3 and theoretical Lidar. My mistake.
Lidar still doesn't work in the rain though.
posted by rhamphorhynchus at 9:35 AM on March 17
The idea the lidars give you a perfect point cloud of the world around you is a myth. I no longer work in autonomous driving but I spent 4 years working on lidar sensors (not for Tesla, obviously) for onroad vehicles.
Lidar does have many strengths but is also subject to the equivalent of optical illusions. The rays can bounce off shiney surfaces (windows, puddles, retroreflective signs) or get bent and reflected in weird ways. They are also very badly affected by weather - even quite mild rain and mist can block the rays and forget about heavy snow. The sensors themselves are rather fragile, although this has improved with the newer solid-state models.
I also think that people underestimate how good the newer vision algorithms are for determining objects - they are incredibly good at picking out objects. I've seen logs where a vehicle drove up to an intersection where a crowd of 300 people milled around and the vehicle tracked them all.
Musks problem is that although humans can drive with only one eye (binocular vision doesn't really count at the distances useful for driving), even expensive cameras are terrible compared to our eyes. The only way and autonomous vehicle is going to have consistent performance is to have lots of redundancy - that means multiple cameras and multiple modes of detection, which lidar may or may not be part of.
> One interesting detail about Tesla's camera only system is that it doesn't use parallax to determine distance. It uses the identification of the object to determine a distance (that looks like car, but it's tiny so it must be far away). They used to have multiple cameras (short, medium, and long range) but got rid of them. That might prove to be the bigger error.
I hadn't heard this (been out of the industry for a while) but that seems, quite literally, short-sighted.
posted by AndrewStephens at 9:37 AM on March 17 [14 favorites]
Lidar does have many strengths but is also subject to the equivalent of optical illusions. The rays can bounce off shiney surfaces (windows, puddles, retroreflective signs) or get bent and reflected in weird ways. They are also very badly affected by weather - even quite mild rain and mist can block the rays and forget about heavy snow. The sensors themselves are rather fragile, although this has improved with the newer solid-state models.
I also think that people underestimate how good the newer vision algorithms are for determining objects - they are incredibly good at picking out objects. I've seen logs where a vehicle drove up to an intersection where a crowd of 300 people milled around and the vehicle tracked them all.
Musks problem is that although humans can drive with only one eye (binocular vision doesn't really count at the distances useful for driving), even expensive cameras are terrible compared to our eyes. The only way and autonomous vehicle is going to have consistent performance is to have lots of redundancy - that means multiple cameras and multiple modes of detection, which lidar may or may not be part of.
> One interesting detail about Tesla's camera only system is that it doesn't use parallax to determine distance. It uses the identification of the object to determine a distance (that looks like car, but it's tiny so it must be far away). They used to have multiple cameras (short, medium, and long range) but got rid of them. That might prove to be the bigger error.
I hadn't heard this (been out of the industry for a while) but that seems, quite literally, short-sighted.
posted by AndrewStephens at 9:37 AM on March 17 [14 favorites]
As to the question of why this video wasn’t made earlier, they removed the other sensors in 2021-2023.
posted by adamsc at 9:41 AM on March 17 [3 favorites]
posted by adamsc at 9:41 AM on March 17 [3 favorites]
I've been thinking that an effective, legal Tesla protest would be for a few thousand people to "test drive" Teslas each week, simultaneously tying up virtually their entire dealer inventory, wasting staff time, and putting some miles on the vehicles. Any one person wouldn't be able to test drive more than a couple of vehicles, so there would have to be a steady stream of test drivers, but it could make a further dent in sales.
posted by jedicus at 10:01 AM on March 17 [1 favorite]
posted by jedicus at 10:01 AM on March 17 [1 favorite]
Lidar still doesn't work in the rain though.
It did in this case. The car with lidar did fine in the simulated rain storm, whereas the Tesla failed (link to timestamp in the video).
posted by jedicus at 10:03 AM on March 17 [8 favorites]
It did in this case. The car with lidar did fine in the simulated rain storm, whereas the Tesla failed (link to timestamp in the video).
posted by jedicus at 10:03 AM on March 17 [8 favorites]
Lidar still doesn't work in the rain though.
This is interesting, because two of the tests (that LiDAR passed and Tesla did not) were fog and water. The tech used is by Luminar (I think. I don't recall him saying it in the video, but you can see the name on the hood of the car) which uses 1550nm vs 905nm for "standard" LiDAR, but I don't think that gives you any better penetration in the rain.
Curious.
posted by It's Never Lurgi at 10:06 AM on March 17 [2 favorites]
This is interesting, because two of the tests (that LiDAR passed and Tesla did not) were fog and water. The tech used is by Luminar (I think. I don't recall him saying it in the video, but you can see the name on the hood of the car) which uses 1550nm vs 905nm for "standard" LiDAR, but I don't think that gives you any better penetration in the rain.
Curious.
posted by It's Never Lurgi at 10:06 AM on March 17 [2 favorites]
an effective, legal Tesla protest would be for a few thousand people to "test drive" Teslas each week
I thought the dealerships only take orders via a laptop in the corner, to avoid all the nonsense that comes with operating an auto dealer in a lot of states. There is no inventory.
posted by JoeZydeco at 10:12 AM on March 17
I thought the dealerships only take orders via a laptop in the corner, to avoid all the nonsense that comes with operating an auto dealer in a lot of states. There is no inventory.
posted by JoeZydeco at 10:12 AM on March 17
It did in this case
Did they say it worked? It looked like it just stopped because of the wall (of water) in front of it.
posted by rhamphorhynchus at 10:17 AM on March 17
Did they say it worked? It looked like it just stopped because of the wall (of water) in front of it.
posted by rhamphorhynchus at 10:17 AM on March 17
It's Never Lurgi he explains using the autopilot mode after the Tesla failed the first test relying on the auto braking. Which I pretty strongly think it shouldn't fail. It's his own Tesla and FSD isn't an option all of them.
posted by zenon at 10:22 AM on March 17 [4 favorites]
posted by zenon at 10:22 AM on March 17 [4 favorites]
I liked how they had the big foam block wall pre-scored so that any car crashing through would leave an appropriate sharp-edged cartoon “ka-blam” hole in it.
posted by Thorzdad at 10:53 AM on March 17 [5 favorites]
posted by Thorzdad at 10:53 AM on March 17 [5 favorites]
Tomas Pueyo is bullish on Tesla at least when it comes to robo-taxis.
posted by gottabefunky at 11:02 AM on March 17
posted by gottabefunky at 11:02 AM on March 17
Tomas Pueyo is bullish on Tesla at least when it comes to robo-taxis.
I'd never heard of Tomas Pueyo before today but after reading that I've learned he's most definitely a tool.
posted by RonButNotStupid at 11:11 AM on March 17 [12 favorites]
I'd never heard of Tomas Pueyo before today but after reading that I've learned he's most definitely a tool.
posted by RonButNotStupid at 11:11 AM on March 17 [12 favorites]
I hope someone drops a cartoon anvil on Elon Musk and no one opts to peel his body up and re-inflate his flattened corpse.. but..
I think self driving cars are probably safer than human driven cars ... individually.. though we have no way to know since they are basically an unregulated industry!! But even if they are, the bigger problem to safety and survival of humans and the planet is cars themselves.
I think our organizing efforts should be around de-car-ifying cities not on maintaining human drivers. Building high speed trains, making cities denser, trollies, bike lanes, more and more frequent busses, car-free roads etc is safer than any car can ever be on every level from collisions to pollution to climate.
posted by latkes at 11:30 AM on March 17 [6 favorites]
I think self driving cars are probably safer than human driven cars ... individually.. though we have no way to know since they are basically an unregulated industry!! But even if they are, the bigger problem to safety and survival of humans and the planet is cars themselves.
I think our organizing efforts should be around de-car-ifying cities not on maintaining human drivers. Building high speed trains, making cities denser, trollies, bike lanes, more and more frequent busses, car-free roads etc is safer than any car can ever be on every level from collisions to pollution to climate.
posted by latkes at 11:30 AM on March 17 [6 favorites]
The tech used is by Luminar (I think. I don't recall him saying it in the video, but you can see the name on the hood of the car) which uses 1550nm vs 905nm for "standard" LiDAR, but I don't think that gives you any better penetration in the rain.
The big thing is that you can run a 1550nm laser at an order of magnitude or more power than a 905nm because the vitreous humor greatly attenuates transmission of light after 900nm and won't transmit light longer than ~1400nm. Not to mention the retinas cells can't absorb 1550nm light either. So while 1550nm has a greater amount of water absorption you get it back by running at a much higher power with much longer pulses while also diverging far less. Because of the power increase you also get way longer ranges in dry weather for free.
The obvious disadvantage is that regular silicon has a bandgap of 1.12eV and without coaxing doesn't really absorb light with a wavelength greater than 1100nm. So you need more exotic materials for the sensors which means more cost.
posted by Your Childhood Pet Rock at 11:32 AM on March 17 [5 favorites]
The big thing is that you can run a 1550nm laser at an order of magnitude or more power than a 905nm because the vitreous humor greatly attenuates transmission of light after 900nm and won't transmit light longer than ~1400nm. Not to mention the retinas cells can't absorb 1550nm light either. So while 1550nm has a greater amount of water absorption you get it back by running at a much higher power with much longer pulses while also diverging far less. Because of the power increase you also get way longer ranges in dry weather for free.
The obvious disadvantage is that regular silicon has a bandgap of 1.12eV and without coaxing doesn't really absorb light with a wavelength greater than 1100nm. So you need more exotic materials for the sensors which means more cost.
posted by Your Childhood Pet Rock at 11:32 AM on March 17 [5 favorites]
(Is no one going to remark on the fact that he mapped the inside of Space Mountain and then 3D-printed a scaled-down replica of it?! Goddamn, Mark Rober is so much fun!)
posted by wenestvedt at 11:33 AM on March 17 [13 favorites]
posted by wenestvedt at 11:33 AM on March 17 [13 favorites]
The other thing is that Tesla removed the 77GHz radar which is what almost all other cars use for the AEB and active cruise control stuff. It's pretty damn reliable and works in the rain.
(Is no one going to remark on the fact that he mapped the inside of Space Mountain and then 3D-printed a scaled-down replica of it?! Goddamn, Mark Rober is so much fun!)
Rober acts like it's some sort of covert exfiltration but it's not secret. You can see what Space Mountain looks like on the inside whenever there's a problem with the ride and they turn the lights on to get the riders out of the ride. There's literally lights on videos of the entire ride on YouTube.
posted by Your Childhood Pet Rock at 11:38 AM on March 17 [8 favorites]
(Is no one going to remark on the fact that he mapped the inside of Space Mountain and then 3D-printed a scaled-down replica of it?! Goddamn, Mark Rober is so much fun!)
Rober acts like it's some sort of covert exfiltration but it's not secret. You can see what Space Mountain looks like on the inside whenever there's a problem with the ride and they turn the lights on to get the riders out of the ride. There's literally lights on videos of the entire ride on YouTube.
posted by Your Childhood Pet Rock at 11:38 AM on March 17 [8 favorites]
> The other thing is that Tesla removed the 77GHz radar which is what almost all other cars use for the AEB and active cruise control stuff. It's pretty damn reliable and works in the rain.
I don't know how to interpret Tesla going all in on the "cameras only" strategy for self driving as anything other than evidence that they don't believe self driving will ever be commercially available in consumer vehicles, and so they've committed to continue Musk's hucksterism around it, selling the idea of it to the market and their customers as a way to increase profits. (Apple abandoning their car is more proof of this to me, that the smartest people working in this sector just don't think it's gonna happen). Like if Waymo, who is probably doing the best right now, or some other competitor, were able to put a full self driving "level 5" automated car on the market tomorrow, it would devastate Tesla's stock which is soaring almost entirely on the promise that they will transform transportation by being the first to market with full self driving. And yet, they have decided that they'll get there by greatly restricting the tools at their disposal by abandoning both lidar and radar. My *Subaru* would have stopped at that wall with emergency breaking.
posted by dis_integration at 12:13 PM on March 17 [4 favorites]
I don't know how to interpret Tesla going all in on the "cameras only" strategy for self driving as anything other than evidence that they don't believe self driving will ever be commercially available in consumer vehicles, and so they've committed to continue Musk's hucksterism around it, selling the idea of it to the market and their customers as a way to increase profits. (Apple abandoning their car is more proof of this to me, that the smartest people working in this sector just don't think it's gonna happen). Like if Waymo, who is probably doing the best right now, or some other competitor, were able to put a full self driving "level 5" automated car on the market tomorrow, it would devastate Tesla's stock which is soaring almost entirely on the promise that they will transform transportation by being the first to market with full self driving. And yet, they have decided that they'll get there by greatly restricting the tools at their disposal by abandoning both lidar and radar. My *Subaru* would have stopped at that wall with emergency breaking.
posted by dis_integration at 12:13 PM on March 17 [4 favorites]
"That freeway fully self-driving car idea could only be cooked up by a toon"
posted by RonButNotStupid at 12:19 PM on March 17 [2 favorites]
posted by RonButNotStupid at 12:19 PM on March 17 [2 favorites]
Tomas Pueyo is bullish on Tesla at least when it comes to robo-taxis.
From the article:
One day, you’ll hear from a friend who jettisoned her car.
“How do you commute to work?”
“With robotaxis! I ran the math and it’s actually cheaper!”
Commutes will be the key milestone: They represent a huge number of miles because they happen twice a day, every week day, 40-50 weeks per year. Replacing them will accelerate the takeover of robotaxis. How will that happen?
Man, if Robotaxis can replace everyone’s daily commute in a major city it means public transit has well and truly failed in a way that is reprehensible.
posted by Going To Maine at 12:27 PM on March 17 [14 favorites]
From the article:
One day, you’ll hear from a friend who jettisoned her car.
“How do you commute to work?”
“With robotaxis! I ran the math and it’s actually cheaper!”
Commutes will be the key milestone: They represent a huge number of miles because they happen twice a day, every week day, 40-50 weeks per year. Replacing them will accelerate the takeover of robotaxis. How will that happen?
Man, if Robotaxis can replace everyone’s daily commute in a major city it means public transit has well and truly failed in a way that is reprehensible.
posted by Going To Maine at 12:27 PM on March 17 [14 favorites]
In response to an article discussing Rober's video A group of Tesla fans and investors has inadvertently exposed Tesla’s shadiness regarding crashes involving Autopilot by attempting to claim that the advanced driver-assist system was not active in a crash test.
It seems autopilot disengages immediately prior to an unavoidable crash and may have led to claims that autopilot wasn't engaged in crashes where it was actually in control prior to the crash.
posted by NailsTheCat at 12:32 PM on March 17 [8 favorites]
It seems autopilot disengages immediately prior to an unavoidable crash and may have led to claims that autopilot wasn't engaged in crashes where it was actually in control prior to the crash.
posted by NailsTheCat at 12:32 PM on March 17 [8 favorites]
I enjoyed the video very much, though I tend to make a face when someone uses the idea of a kid getting hit by a car as a punchline, even a crash test dummy kid.
As someone who gave up my car a couple years ago, and relies almost entirely on public transportation (I hired a car once in the past couple of years, got a friend to drive me to the dump in her pickup once), I am perplexed about the need for individual autonomous vehicles of any sort except under very limited circumstances *if* (BIG if) we had good public transportation in every area of the country, plus good bicycle lanes.
I know people's problems with the idea. They range from having lots of kids to living in areas unserved by public transportation to needing transportation because of handicaps, to (the most common in my area, apparently) being white in a majority black city so they're afraid of public transportation. (I am often the only white person on the bus). But so much is wasted in the way of raw materials, engineering, insurance, fuel, and licensing is involved in everyone having one or two vehicles, not to mention the congestion, pollution, and mayhem involved in everyone doodling around in their motorized living rooms. You add self-driving to the mix and it's comical.
posted by Peach at 12:33 PM on March 17 [6 favorites]
As someone who gave up my car a couple years ago, and relies almost entirely on public transportation (I hired a car once in the past couple of years, got a friend to drive me to the dump in her pickup once), I am perplexed about the need for individual autonomous vehicles of any sort except under very limited circumstances *if* (BIG if) we had good public transportation in every area of the country, plus good bicycle lanes.
I know people's problems with the idea. They range from having lots of kids to living in areas unserved by public transportation to needing transportation because of handicaps, to (the most common in my area, apparently) being white in a majority black city so they're afraid of public transportation. (I am often the only white person on the bus). But so much is wasted in the way of raw materials, engineering, insurance, fuel, and licensing is involved in everyone having one or two vehicles, not to mention the congestion, pollution, and mayhem involved in everyone doodling around in their motorized living rooms. You add self-driving to the mix and it's comical.
posted by Peach at 12:33 PM on March 17 [6 favorites]
Tesla is also a ridiculously overvalued company whose hyperinflated share price only exists because investors love to throw cash at Musk. Tesla's executives know this, which is why they've been selling off their overvalued stocks to average American's 401ks for ages now. When the Tesla stock bubble bursts, Musk will be fine. He has huge amounts of money tied up in SpaceX and Starlink, and that government contract firehose isn't stopping anytime soon. At worst, Musk will stay downgraded from "richest human in history" to "elite rich billionaire," which is hardly a punishment. It's shitty to think he'll never face any accountability for his naked corruption, but that seems to be the world we're living in.
posted by lock robster at 12:43 PM on March 17 [3 favorites]
posted by lock robster at 12:43 PM on March 17 [3 favorites]
I'm surprised at all the apparently knowledgeable folks pointing out that LIDAR has weaknesses as if anyone suggested using LIDAR alone.
The ridiculously high benefit:cost ratio of visual light cameras makes it pretty stupid not to include them. They're practically free next to LIDAR, and provide a crazy amount of information -- except for depth. Of course the best solution is LIDAR + RADAR + cameras + sensor fusion, with each sensor's strength compensating for the weaknesses of the other.
posted by CaseyB at 12:44 PM on March 17 [5 favorites]
The ridiculously high benefit:cost ratio of visual light cameras makes it pretty stupid not to include them. They're practically free next to LIDAR, and provide a crazy amount of information -- except for depth. Of course the best solution is LIDAR + RADAR + cameras + sensor fusion, with each sensor's strength compensating for the weaknesses of the other.
posted by CaseyB at 12:44 PM on March 17 [5 favorites]
I seem to remember there were possible interference issues with lidar if multiple are operating the same vicinity.
I'm no expert on LIDAR, but the same thing was once the case with RADAR. It was solved pretty quickly due to the, uh, highly adversarial environment it was developed in (World War II), but I'd think the same solutions would work for LIDAR especially with the additional bandwidth you get at the much higher frequencies.
Basically, you don't just "believe" any signal you get back at your receiver is genuine; instead, you transmit a pulse train that contains some sort of unique code (typically a convolution code resulting in a spread-spectrum signal) and then you look for reflections that match (or are distortions of) that code, and only that code. Looking at the time between transmit and arrival gives you the distance to the target, while looking at the stretching or compressing of the code (redshift/blueshift) gives you the target's relative motion.
I'd honestly be a bit surprised if modern LIDAR systems don't already do something like that? But maybe if they are coming directly from situations like indoor robotics, they don't include it since it's unlikely that another LIDAR emitter is operating in the same vicinity. But for vehicle use in an uncontrolled environment you'd be pretty dumb not to do it.
I was at one point pretty enthusiastic about the possibilities of self-driving cars, because honestly people aren't even that great at driving cars, so the goal for an autonomy system isn't—and shouldn't be—absolute perfection, but just being better than a human. And humans operate solely using visible light, in some cases with highly-compromised depth perception, with only a small high-resolution visual field and a ton of interpolation done by your brain to piece together your coherent visual picture. It's not obviously out of the realm of possibility that you could produce a computer system that could do better.
However, I've mostly soured on the concept not for technical reasons, but implementation-related ones. I don't think the companies that are trying to do autonomy are set up for success, and I think they are highly incentivized to cut corners. That's a recipe for disaster, and I don't trust any system that has a for-profit, investor-backed company behind it. It's just too easy to smoke-and-mirrors your way through a bunch of closed-course demos and then fail in real-world edge cases.
Plus, as the automatic systems get better and better, those same sensors and technologies will be applied to driver-augmentation systems that make a human even better, so the criteria for success isn't static. "Better than a human" is a moving target; a human driver in a 1995 car isn't the same as a human driver in a 2025 model year one, and it won't be the same as a 2035 car either (assuming we aren't scratching around in the ruins of civilization by 2035). At some point we'll probably have automatic collision avoidance standard on all human-driven cars, just like we have turn signals and airbags.
Lots of the "tricks" that self-driving cars allow for, like optimization of stoplights to allow nonstop motion through intersections, could be done by aided human drivers, in the same way that instrument landing systems on aircraft allow pilots to land planes in situations where their eyeballs wouldn't let them do it safely unaided. But those systems don't replace pilots, they augment them. I think that's probably where all the "self driving" technology will eventually go.
posted by Kadin2048 at 12:48 PM on March 17 [7 favorites]
I'm no expert on LIDAR, but the same thing was once the case with RADAR. It was solved pretty quickly due to the, uh, highly adversarial environment it was developed in (World War II), but I'd think the same solutions would work for LIDAR especially with the additional bandwidth you get at the much higher frequencies.
Basically, you don't just "believe" any signal you get back at your receiver is genuine; instead, you transmit a pulse train that contains some sort of unique code (typically a convolution code resulting in a spread-spectrum signal) and then you look for reflections that match (or are distortions of) that code, and only that code. Looking at the time between transmit and arrival gives you the distance to the target, while looking at the stretching or compressing of the code (redshift/blueshift) gives you the target's relative motion.
I'd honestly be a bit surprised if modern LIDAR systems don't already do something like that? But maybe if they are coming directly from situations like indoor robotics, they don't include it since it's unlikely that another LIDAR emitter is operating in the same vicinity. But for vehicle use in an uncontrolled environment you'd be pretty dumb not to do it.
I was at one point pretty enthusiastic about the possibilities of self-driving cars, because honestly people aren't even that great at driving cars, so the goal for an autonomy system isn't—and shouldn't be—absolute perfection, but just being better than a human. And humans operate solely using visible light, in some cases with highly-compromised depth perception, with only a small high-resolution visual field and a ton of interpolation done by your brain to piece together your coherent visual picture. It's not obviously out of the realm of possibility that you could produce a computer system that could do better.
However, I've mostly soured on the concept not for technical reasons, but implementation-related ones. I don't think the companies that are trying to do autonomy are set up for success, and I think they are highly incentivized to cut corners. That's a recipe for disaster, and I don't trust any system that has a for-profit, investor-backed company behind it. It's just too easy to smoke-and-mirrors your way through a bunch of closed-course demos and then fail in real-world edge cases.
Plus, as the automatic systems get better and better, those same sensors and technologies will be applied to driver-augmentation systems that make a human even better, so the criteria for success isn't static. "Better than a human" is a moving target; a human driver in a 1995 car isn't the same as a human driver in a 2025 model year one, and it won't be the same as a 2035 car either (assuming we aren't scratching around in the ruins of civilization by 2035). At some point we'll probably have automatic collision avoidance standard on all human-driven cars, just like we have turn signals and airbags.
Lots of the "tricks" that self-driving cars allow for, like optimization of stoplights to allow nonstop motion through intersections, could be done by aided human drivers, in the same way that instrument landing systems on aircraft allow pilots to land planes in situations where their eyeballs wouldn't let them do it safely unaided. But those systems don't replace pilots, they augment them. I think that's probably where all the "self driving" technology will eventually go.
posted by Kadin2048 at 12:48 PM on March 17 [7 favorites]
A consortium of basically everyone-but-Tesla* is building out a competitive network called Ionna.
Meanwhile, a large number of EVs from other manufacturers are now able to fast-charge via Tesla Superchargers. Newer Tesla Superchargers even include a "Magic Dock" that allow them to be used with compatible non-Tesla cars without the need for an adapter, although the adapter isn't an exorbitant expense. Ford, GM, Lucid, Mercedes, Nissan, Rivian and Volvo/Polestar EVs can charge at Tesla Superchargers in addition to non-Tesla CCS chargers or Level 1 or 2 destination chargers.
posted by eschatfische at 12:49 PM on March 17 [1 favorite]
Meanwhile, a large number of EVs from other manufacturers are now able to fast-charge via Tesla Superchargers. Newer Tesla Superchargers even include a "Magic Dock" that allow them to be used with compatible non-Tesla cars without the need for an adapter, although the adapter isn't an exorbitant expense. Ford, GM, Lucid, Mercedes, Nissan, Rivian and Volvo/Polestar EVs can charge at Tesla Superchargers in addition to non-Tesla CCS chargers or Level 1 or 2 destination chargers.
posted by eschatfische at 12:49 PM on March 17 [1 favorite]
I mean, Tesla is being paid for the privilege of cats using their network.
posted by Going To Maine at 12:52 PM on March 17
posted by Going To Maine at 12:52 PM on March 17
the big foam block wall pre-scored so that any car crashing through would leave an appropriate sharp-edged cartoon “ka-blam” hole in it.
Sheets of foam often come in 24" wide sizes.
But "better-than-human on average" is a low bar.
And one that is unlikely to persist going forward. Tesla already proved we don't care if their cars have worse fire fatality stas than the Pinto. And they allow owners to dial in how much they want to break the law:
Even if self driving cars manage to be consistently safer than people going forward I expect, short of regulation that looks unlikely, that their numbers will slowly get worse. America is willing to sacrifice 40k lives a year to car culture, I can't imagine Musk et. al. ignoring that if it sells more cars.
posted by Mitheral at 12:54 PM on March 17 [6 favorites]
Sheets of foam often come in 24" wide sizes.
But "better-than-human on average" is a low bar.
And one that is unlikely to persist going forward. Tesla already proved we don't care if their cars have worse fire fatality stas than the Pinto. And they allow owners to dial in how much they want to break the law:
Thousands of Teslas are now being equipped with a feature that prompts the car to break common traffic laws — and the revelation is prompting some advocates to question the safety benefits of automated vehicle technology when unsafe human drivers are allowed to program it to do things that endanger other road users.And I can't find the article but waymo is apparently modifying their software to be more aggressive to intimidate pedestrians. Because waiting as legally required increases trip time.
Even if self driving cars manage to be consistently safer than people going forward I expect, short of regulation that looks unlikely, that their numbers will slowly get worse. America is willing to sacrifice 40k lives a year to car culture, I can't imagine Musk et. al. ignoring that if it sells more cars.
posted by Mitheral at 12:54 PM on March 17 [6 favorites]
I'm surprised at all the apparently knowledgeable folks pointing out that LIDAR has weaknesses as if anyone suggested using LIDAR alone.
My objection was that lidar fails in enough of the circumstances in which you'd expect a car to operate that any self driving system has to work without it. In which case, why include it? Your Childhood Pet Rock's comment raises some reasonable questions about whether I'm still correct about that, but the video in the post shows the point cloud completely dissolving in the "rain" for all that it also shows the car stopping.
posted by rhamphorhynchus at 1:45 PM on March 17 [1 favorite]
My objection was that lidar fails in enough of the circumstances in which you'd expect a car to operate that any self driving system has to work without it. In which case, why include it? Your Childhood Pet Rock's comment raises some reasonable questions about whether I'm still correct about that, but the video in the post shows the point cloud completely dissolving in the "rain" for all that it also shows the car stopping.
posted by rhamphorhynchus at 1:45 PM on March 17 [1 favorite]
I mean, Tesla is being paid for the privilege of cats using their network.
Well then it's just Toonces, really, I hardly think that's worth much consideration.
posted by axiom at 2:47 PM on March 17 [9 favorites]
Well then it's just Toonces, really, I hardly think that's worth much consideration.
posted by axiom at 2:47 PM on March 17 [9 favorites]
> why wouldn't you want your machines of the future to have a vast and interesting sensorium?
Elon Musk thinking: his kids are important. Your kids are targets.
posted by JustSayNoDawg at 3:26 PM on March 17 [2 favorites]
Elon Musk thinking: his kids are important. Your kids are targets.
posted by JustSayNoDawg at 3:26 PM on March 17 [2 favorites]
It's not particularly important and I'm way down thread but calling mark rober a tech bro dude is a pretty hit take. He's an openly progressive guy, has made a ton of videos and comments on the environment, science, etc and is, as far as you can ever know a creator, a very solid dude. Anyone who'd go to that length for squirrels is a decent person.
posted by chasles at 4:01 PM on March 17 [6 favorites]
posted by chasles at 4:01 PM on March 17 [6 favorites]
Lidar still doesn't work in the rain though.
It did in this case. The car with lidar did fine in the simulated rain storm, whereas the Tesla failed (link to timestamp in the video).
Did they say it worked? It looked like it just stopped because of the wall (of water) in front of it.
They did claim the LIDAR worked (ie identified the child and stopped before hitting it) in the rain, but I think they're at least partly wrong. It looked to me as if the LIDAR detected the rain itself as an object, although not perfectly because it didn't stop before the wall of water. It stopped a long way back from the simulated child and just slightly within the rain area. A more accurate test would be whether it could detect and navigate around an object without stopping under those circumstances.
Autonomous braking is now required on all new cars here in Australia. Given the average age of cars in Australia is 10.6 years, it's going to be a long time before even a majority of cars have this functionality and many countries even longer (including the US). I remain of the view that mixing autonomous and human drivers presents significant challenges, so I don't expect to see autonomous cars as anything more than a 'here is the future' gimmick in my lifetime. I do agree that technology can enhance the ability of human drivers, but suspect there is a point where they just stop paying attention rather than using the technology to assist them.
I was impressed that they cut the foam in the wall to make that perfect cartoon burst, but also that they printed the back of the wall with bricks to make the effect just perfect from both sides.
posted by dg at 4:17 PM on March 17 [2 favorites]
It did in this case. The car with lidar did fine in the simulated rain storm, whereas the Tesla failed (link to timestamp in the video).
Did they say it worked? It looked like it just stopped because of the wall (of water) in front of it.
They did claim the LIDAR worked (ie identified the child and stopped before hitting it) in the rain, but I think they're at least partly wrong. It looked to me as if the LIDAR detected the rain itself as an object, although not perfectly because it didn't stop before the wall of water. It stopped a long way back from the simulated child and just slightly within the rain area. A more accurate test would be whether it could detect and navigate around an object without stopping under those circumstances.
Autonomous braking is now required on all new cars here in Australia. Given the average age of cars in Australia is 10.6 years, it's going to be a long time before even a majority of cars have this functionality and many countries even longer (including the US). I remain of the view that mixing autonomous and human drivers presents significant challenges, so I don't expect to see autonomous cars as anything more than a 'here is the future' gimmick in my lifetime. I do agree that technology can enhance the ability of human drivers, but suspect there is a point where they just stop paying attention rather than using the technology to assist them.
I was impressed that they cut the foam in the wall to make that perfect cartoon burst, but also that they printed the back of the wall with bricks to make the effect just perfect from both sides.
posted by dg at 4:17 PM on March 17 [2 favorites]
You don't need to go full looney toons to get a Tesla to crash into something, just put a couple of small stickers on the ground.
posted by stilgar at 6:22 PM on March 17
posted by stilgar at 6:22 PM on March 17
Rober acts like it's some sort of covert exfiltration but it's not secret. You can see what Space Mountain looks like on the inside whenever there's a problem with the ride and they turn the lights on to get the riders out of the ride.
He's upfront that he's wanted to use some tech wizardy gadget to 'map' the inside of space mountain for himself since he was a kid. The portable lidar sensor was a way to do that. And using that information to build a highly accurate 3d printed model is just gravy. I'm sure he's seen lights-on videos, and it would have been easier to just wear NV goggles, but this was much cooler.
The sneaking about was because Disney park security (including the undercover ones) are no joke when you start doing suspicious things like carrying a big backpack of electronics with lots of flashy IR lights or looking too much like an unauthorised film crew - not because the space mountain ride itself is some military-grade secret.
I dunno, it seemed like fulfilling a childhood dream was the real purpose, but he decided to add some extra lidar stuff on to see how it compared to his tesla. And given the state of the world right now, Rober geeking out on something harmless was a nice 10 minutes, with some bonus ragging on tesla's AEB inadequacies as a treat. Getting the Musk stans/nazis frothing at the mouth over an 'unfair test' (AEB should be working entirely independently of any cruise control/lane keeping system, and it should work a shit-ton better than demonstrated if they want to stay road legal in the EU/UK) is an extra extra bonus!
posted by Absolutely No You-Know-What at 7:05 PM on March 17 [2 favorites]
He's upfront that he's wanted to use some tech wizardy gadget to 'map' the inside of space mountain for himself since he was a kid. The portable lidar sensor was a way to do that. And using that information to build a highly accurate 3d printed model is just gravy. I'm sure he's seen lights-on videos, and it would have been easier to just wear NV goggles, but this was much cooler.
The sneaking about was because Disney park security (including the undercover ones) are no joke when you start doing suspicious things like carrying a big backpack of electronics with lots of flashy IR lights or looking too much like an unauthorised film crew - not because the space mountain ride itself is some military-grade secret.
I dunno, it seemed like fulfilling a childhood dream was the real purpose, but he decided to add some extra lidar stuff on to see how it compared to his tesla. And given the state of the world right now, Rober geeking out on something harmless was a nice 10 minutes, with some bonus ragging on tesla's AEB inadequacies as a treat. Getting the Musk stans/nazis frothing at the mouth over an 'unfair test' (AEB should be working entirely independently of any cruise control/lane keeping system, and it should work a shit-ton better than demonstrated if they want to stay road legal in the EU/UK) is an extra extra bonus!
posted by Absolutely No You-Know-What at 7:05 PM on March 17 [2 favorites]
How does that make [Tesla] different from those [Musk] rails against?
Tesla: We have cameras.
posted by stopgap at 7:12 PM on March 17
Tesla: We have cameras.
posted by stopgap at 7:12 PM on March 17
You can see what Space Mountain looks like on the inside whenever there's a problem with the ride and they turn the lights on
TIL that Space Mountain is just a roller coaster inside with the lights off.
posted by 922257033c4a0f3cecdbd819a46d626999d1af4a at 7:26 PM on March 17 [1 favorite]
TIL that Space Mountain is just a roller coaster inside with the lights off.
posted by 922257033c4a0f3cecdbd819a46d626999d1af4a at 7:26 PM on March 17 [1 favorite]
Golly, I wonder why this is only coming out now. It's not like they haven't been using a problematic vision-only solution for years.
I can answer that.
What's problematic isn't the vision only solution. It's how it's implemented.
The problem isn't that the Tesla couldn't detect the child in the fog or rain at 40 MPH.
It's that the Tesla stayed at 40 MPH with fog and rain ahead.
A self driving car can detect when its own detectors are compromised by the environment.
The decision to maintain speed when it needs longer has the right stopping distance is embedded in code written by chuds. The same code, if written by actual engineers, would have forced the car to slow down long beforehand.
posted by ocschwar at 8:14 PM on March 17 [9 favorites]
I can answer that.
What's problematic isn't the vision only solution. It's how it's implemented.
The problem isn't that the Tesla couldn't detect the child in the fog or rain at 40 MPH.
It's that the Tesla stayed at 40 MPH with fog and rain ahead.
A self driving car can detect when its own detectors are compromised by the environment.
The decision to maintain speed when it needs longer has the right stopping distance is embedded in code written by chuds. The same code, if written by actual engineers, would have forced the car to slow down long beforehand.
posted by ocschwar at 8:14 PM on March 17 [9 favorites]
Just saying, the squirrel video linked is totally adorable, and worth the 15min to watch.
posted by hovey at 8:18 PM on March 17 [3 favorites]
posted by hovey at 8:18 PM on March 17 [3 favorites]
TIL that Space Mountain is just a roller coaster inside with the lights off.
My mother used to run educational programs at Disney parks. One of the activities was riding Space Mountain twice, once normally, and once with the lights on.
posted by funkaspuck at 4:14 AM on March 18
My mother used to run educational programs at Disney parks. One of the activities was riding Space Mountain twice, once normally, and once with the lights on.
posted by funkaspuck at 4:14 AM on March 18
I find Mark Rober to be annoyingly curated and aggravatingly bland. While other science advocates often attract undue controversy because of the topics they're sharing or even for just existing, Rober always seems to go out of his way to keep things capital-A Apolitical. After all, evangelical parents want their kids to grow up to be engineers too and that's money on the table for Rober's online STEM courses and his robotics kits.
This is kind of why this video fascinates me. Tesla's cavalier attention to safety has been known for a while, but the brand has enough relentless stans to make life miserable for anyone who's going to point it out. Has the political climate changed where Rober feels it's safe or even economically viable to make this sort of video? He's certainly getting a lot of backlash from the usual corners of the internet. Is he actually making the smallest of steps towards having an opinion?
posted by RonButNotStupid at 5:42 AM on March 18
This is kind of why this video fascinates me. Tesla's cavalier attention to safety has been known for a while, but the brand has enough relentless stans to make life miserable for anyone who's going to point it out. Has the political climate changed where Rober feels it's safe or even economically viable to make this sort of video? He's certainly getting a lot of backlash from the usual corners of the internet. Is he actually making the smallest of steps towards having an opinion?
posted by RonButNotStupid at 5:42 AM on March 18
« Older Female Scribes in the Middle Ages | Of Dandelions and Palestine Newer »
posted by Lemkin at 7:35 AM on March 17 [24 favorites]