Tesla Thread

[mikegarrison]

58 minutes ago, tomf said:

Has the autopilot liability been tested in a US court yet? It's hard to imagine that no enterprising lawyer yet hasn't tried an argument that a product that is so easy to use irresponsibly, even with good intentions must give some liability to the manufacturer.

If I were Tesla's lawyers I would be up at night worrying.

There have been multiple lawsuits. Many are currently pending.

[CatastrophicFailure]

57 minutes ago, tomf said:

Has the autopilot liability been tested in a US court yet? It's hard to imagine that no enterprising lawyer yet hasn't tried an argument that a product that is so easy to use irresponsibly, even with good intentions must give some liability to the manufacturer.

If I were Tesla's lawyers I would be up at night worrying.

But you see, it's not. Consumer Reports recently demonstrated how "easy" it was, in that doing such required deliberately ignoring or flat out disabling multiple safety features. There were follow up videos from other sources demonstrating the same behavior from other manufacturers' systems, too. Eye-tracking could be circumvented just with sunglasses on the headrest or even googly eyes. 

[mikegarrison]

Three years ago, Tesla settled a class action suit for supposedly over-promising the safety of Autopilot.

[tomf]

Just now, mikegarrison said:

There have been multiple lawsuits. Many are currently pending.

Any that have gone to trial with a public verdict though?

Ok I can just Google that, I can't find any cases that have publicly concluded, although Germany did fine Tesla for overstating the autopilot capability

[mikegarrison]

1 minute ago, tomf said:

Any that have gone to trial with a public verdict though?

That's part of why companies settle lawsuits. Because that way, there is no public decision against them.

[CatastrophicFailure]

9 minutes ago, mikegarrison said:

That's part of why companies settle lawsuits. Because that way, there is no public decision against them.

Also because fighting a lawsuit is almost always far more expensive, even if you win. 

[tater]

45 minutes ago, mikegarrison said:

That's part of why companies settle lawsuits. Because that way, there is no public decision against them.

I have a feeling the legal hurdles of FSD are pretty non-trivial (vs Autopilot, which is something other cars have a version of as well—cruise control, plus lane-keeping, plus speed control relative to cars in front, etc).

Even if FSD was safer than human drivers, when a human driver clobbers someone, blame is easier to assign, and hence liability. What happens in that FSD world when one of the presumably rare accidents happens?

[tater]

My buddy (just texted him) says he expects it will get litigated to different results in every state.

[benzman]

My niece once took me somewhere in her Golf.  As she didn't  know how to get there she took out her phone, thumbed up a GPS app, dialed in the destination then studied the route.  All while driving and taking the occasional look outside. When I nervously asked her why she didn't get a proper GPS so she could hear directions and take a quick look down at the screen now and again she airily replied that she preferred to look at her phone because she was a visual person, not an auditory person! It all helped to reinforce my opinion that humans in a car are unpredictable loose cannons, and that the sooner self-driving cars become a lot better the sooner the road toll will show a precipitous decline.

On another note, a couple of years ago I took a demonstration ride in a Tesla.  We were going along a multi-lane, slightly winding freeway and I was enjoying the sensation of the car keeping between the white lines, slowing down behind a slower car and then speeding up again when the road was clear. We came up to a truck in the lane to  our left and, just as our front was about level with the truck's rear it started to move into our lane.  I don't know whether the car would have done anything because I immediately turned the auto off and the car slowed down. Thinking about it later it occurred to me that as it was my first experience of auto driving I was alert and had my finger poised over the kill switch.  I wonder how many Tesla crashes have been caused because the driver has been daydreaming, distracted or playing games on his phone?

[DDE]

1 hour ago, benzman said:

When I nervously asked her why she didn't get a proper GPS

Overpriced, inflexible, potentially rarely updated vis-a-vis smartphone apps. Strapping a phone or tablet to the dashboard seems like the easier option.

That said, I've never seen anyone listening to GPS correctly navigate anything more complex than a cloverleaf interchange. Gotta read those maps... at which point, audio becomes a secondary channel.

Edited August 31, 2021 by DDE

[CatastrophicFailure]

1 hour ago, benzman said:

I wonder how many Tesla crashes have been caused because the driver has been daydreaming, distracted or playing games on his phone?

Likely less than a control group of non-Tesla drivers being distracted: 

Quote

“…We registered one accident for every 4.19 million miles driven in which drivers had Autopilot engaged. For those driving without Autopilot but with our active safety features, we registered one accident for every 2.05 million miles driven. For those driving without Autopilot and without our active safety features, we registered one accident for every 978 thousand miles driven. By comparison, NHTSA’s most recent data shows that in the United States there is an automobile crash every 484,000 miles.”

 

28 minutes ago, DDE said:

Overpriced, inflexible, potentially rarely updated vis-a-vis smartphone apps. Strapping a phone or tablet to the dashboard seems like the easier option.

That said, I've never seen anyone listening to GPS correctly navigate anything more complex than a cloverleaf interchange. Gotta read those maps... at which point, audio becomes a secondary channel.

The solution, then, is to integrate the GPS right into the car itself, which every single manufacturer has done to some extent, by now, and the bigger the display screen, the better. Then eventually you take the driver out of the loop entirely. 

Ive had one experience, to this point, driving with Navigate on Autopilot in a strange city. NoA will take you from on-ramp to off-ramp, even handling lane changes, (presumably) with no interaction beyond paying attention and keeping a hand on the wheel. 
 

In every single case where I second-guessed the car and disengaged it… the car was right, and I took the wrong exit. 

[RCgothic]

It isn't hard to imagine a fully self driving vehicle being safer than a human. The self-driving vehicle can see in every direction at once, never disregards traffic laws, never gets bored, distracted, or drivers whilst impaired by fatigue or substances. Humans are the weak link, and the data shows self-driving vehicles are *already* a lot safer than human drivers.

I don't have a Tesla. I have a high-end 2018 Ford with adaptive cruise control and lane keeping.

Ford's 2018 adaptive cruise control is great on long journeys, though it does have flaws. It tends to brake too aggressively when the traffic in front slows down, and leaves a gap to the car ahead that invites other vehicles to cut in front unless you use a short distance setting that aggravates the aggressive braking. It also sometimes gets confused by corners, braking for traffic in other lanes. So in practice I tend to takeover when I notice traffic about to merge into my lane or slowing down ahead, which limits the usefulness a little.

The lane-keeping is definitely best described as a driver-assist. It won't keep the vehicle centred, but applies a correction (and/or steering wheel vibration) if it approaches a line. On a corner it'll correct the first time, but then the car will drift over thereafter. And on straight lanes it will tend to pinball until the angle gets too large to correct. It's not good enough to release the steering wheel, and you'll get a shouty driver-alert to keep hands on steering wheel if it doesn't notice you applying some steering wheel torque for too long (even if your hands are on the wheel). Though I have the lane-keeping on at all times, generally it only gets to kick in if I'm impaired by fatigue (which the car also recognises and warns about), so that's a good time to stop or switch drivers.

I can definitely agree that if my attention weren't required for lane-keeping my attention would not be on the road. I can't spend long periods of time babysitting an automatic function, is just not how my brain works. I have friends who can perform low-activity monotonous tasks for ages, but I'm definitely not one of that sort of people.

[magnemoe]

2 hours ago, CatastrophicFailure said:

Likely less than a control group of non-Tesla drivers being distracted: 

“…We registered one accident for every 4.19 million miles driven in which drivers had Autopilot engaged. For those driving without Autopilot but with our active safety features, we registered one accident for every 2.05 million miles driven. For those driving without Autopilot and without our active safety features, we registered one accident for every 978 thousand miles driven. By comparison, NHTSA’s most recent data shows that in the United States there is an automobile crash every 484,000 miles.”

The solution, then, is to integrate the GPS right into the car itself, which every single manufacturer has done to some extent, by now, and the bigger the display screen, the better. Then eventually you take the driver out of the loop entirely. 

Ive had one experience, to this point, driving with Navigate on Autopilot in a strange city. NoA will take you from on-ramp to off-ramp, even handling lane changes, (presumably) with no interaction beyond paying attention and keeping a hand on the wheel. 

In every single case where I second-guessed the car and disengaged it… the car was right, and I took the wrong exit. 

4, 2 and 1 million km is Tesla's at comparable roads, say the automated systems don't work because of say snow its also an much higher chance of an accident?
The .48 millions makes sense overall, few has Tesla as their first car for one and inexperienced drivers has an much higher accident rate. 

[kerbiloid]

1 hour ago, RCgothic said:

It isn't hard to imagine a fully self driving vehicle being safer than a human.

You are probably living in a place where the human drivers are more wise and lawful that AI, so get the point.

 

1 hour ago, magnemoe said:

4, 2 and 1 million km is Tesla's at comparable roads, say the automated systems don't work because of say snow its also an much higher chance of an accident?

So, Tesla takes the easy part, leaving the human driver without training and letting him take the control in complicated cases immediately.

Kinda it says: "... But you are an expert, human, hehe. Show us your master-class, and rememeber that in any case you are guilty, not me."

Edited August 31, 2021 by kerbiloid

[tater]

6 hours ago, CatastrophicFailure said:

Likely less than a control group of non-Tesla drivers being distracted: 

Yeah, this is obviously required. What are the active safety features on the tesla? Will it stop itself without autopilot on (if car is closing too fast on a vehicle ahead, etc)?

6 hours ago, CatastrophicFailure said:

The solution, then, is to integrate the GPS right into the car itself, which every single manufacturer has done to some extent, by now, and the bigger the display screen, the better. Then eventually you take the driver out of the loop entirely. 

Yeah, our cars have had built in nav systems for years now.

[mikegarrison]

3 hours ago, tater said:

our cars have had built in nav systems for years now.

I drive a manual transmission 2005 Honda Element. It has a radio. That's about it, as far as driver electronics.

[CatastrophicFailure]

3 hours ago, tater said:

Yeah, this is obviously required. What are the active safety features on the tesla? Will it stop itself without autopilot on (if car is closing too fast on a vehicle ahead, etc)?

It's got forward automatic emergency braking. You can set it to warn early, medium, or late, and then it should stop the car before an impact, up to a point. Would probably be good on a lower-speed city street, but I dunno about an interstate. More on that in a sec. I think I've only had actual braking kick on maybe once in the last couple years, I keep the warning set to Early and that's gone off a number of times, usually right about the time my foot is heading for the brake pedal. It'll also automatically reduce power and acceleration in a congested area, like a parking lot, to hopefully reduce those "hit the gas not the brake" incidents. Like @RCgothic's truck, it's also got lane-departure safety, it'll beep, vibrate the wheel, or physically move you back into the lane even if Autopilot is not enabled. And the usual round of blind spot warnings and such. One thing it definitely lacks, which others do have, is rear-cross-traffic alerts.

Tesla has had to recently "recertify" it's auto-emergency braking software, as they're transitioning from radar+vision to a pure vision approach. I've read lots of back-and-forth on this, and there's a compelling argument for pure vision, but it is harder, that's why there haven't' been a lot of updates until recently.  I'm wondering how many of these "hit a thing on the side of the road" type incidents happened because of confusion between radar and vision; if the radar says one thing and the cameras say another, which one should the car beleive? From what I've heard from others, the pure-vision approach does indeed seem to be working better overall, and has drastically reduced if not eliminated the obnoxious phantom braking. I can't speak to that myself, I'm 99% sure my Model X is still using radar. I think the wife's 3 has the newer vision-only software, but I haven't really driven it in months to compare.

8 hours ago, magnemoe said:

4, 2 and 1 million km is Tesla's at comparable roads, say the automated systems don't work because of say snow its also an much higher chance of an accident?
The .48 millions makes sense overall, few has Tesla as their first car for one and inexperienced drivers has an much higher accident rate. 

And that's where it gets tricky, since Autopilot can actually work in snow if the ruts are apparent enough, but it will shut down in really heavy rain. This is the only data available, AFAIK, but the point that "Autopilot is already safer than the average human driver" remains true even if Tesla's data is off by a factor of almost 9.

7 hours ago, kerbiloid said:

So, Tesla takes the easy part, leaving the human driver without training and letting him take the control in complicated cases immediately.

Kinda it says: "... But you are an expert, human, hehe. Show us your master-class, and rememeber that in any case you are guilty, not me."

This is how evolution works. It's literally the only way make a functional system, there is no other way to gain the billions of data-miles to make autonomy work. If "you" are uncomfortable with that, then don't use it. If you do choose to use it, then you accept the responsibility that comes along with that.

40 minutes ago, mikegarrison said:

I drive a manual transmission 2005 Honda Element. It has a radio. That's about it, as far as driver electronics.

I'm old.

I drive for a living. I've spent decades now wrenching on everything from motorcycles to big, loud diesels, sports cars, lawn mowers. I'm more than happy to finally have a ride that requires almost no maintenance from me, and will be quite happy when it can also drive me 95% of the way to work, so I can do drive someone else to work, knowing full well that sooner or later it's gonna put me out of a job, too.

[kerbiloid]

5 minutes ago, CatastrophicFailure said:

I drive for a living. I've spent decades now wrenching on everything from motorcycles to big, loud diesels, sports cars, lawn mowers. I'm more than happy to finally have a ride that requires almost no maintenance from me, and will be quite happy when it can also drive me 95% of the way to work

And that's the difference. You are driving for decades. Like a dolphin, with brain hemispheres sleeping in turns, lol. So, an autopilot is an assistance for you.

While most part of autopilot users are amateurs, which can't be simultaneously alert and sleeping at once.

[tater]

1 hour ago, mikegarrison said:

I drive a manual transmission 2005 Honda Element. It has a radio. That's about it, as far as driver electronics.

The BMW is manual transmission, but it still has a nav system. Before that my Saabs had a radio, lol. The later one had a CD player!

Both have hands-free on the phone, which is nice, though I'm not one to talk on the phone in the car much, 99% of the use is my wife calling from her car to mine in case there's any logistics that needs to be done before we get where we are going—the lion's share over in a few seconds, which is nice from a button on the steering wheel (even does voice recognition).

35 minutes ago, CatastrophicFailure said:

And that's where it gets tricky, since Autopilot can actually work in snow if the ruts are apparent enough, but it will shut down in really heavy rain. This is the only data available, AFAIK, but the point that "Autopilot is already safer than the average human driver" remains true even if Tesla's data is off by a factor of almost 9.

Watching their new approach on the AI day... it's kind of amazing. I'm one of those who tends to think that actual FSD is a lot closer to an "AI complete" problem than many imagine, and now I'm starting to wonder if narrow AI might literally come along for the ride sooner rather than later (in the Dojo space, not in each car, obviously, not nearly enough computer there).

[StrandedonEarth]

Re radar vs vision: As I understand radar, it sends out microwaves which reflect back to the sensor. If it hits an angled metal surface (like the top of an overturned semi) most of the radar signal will reflect away from the car sending the signal, so the radar won’t “see” it. Unlike light, which scatters in all directions (mirrored surfaces notwithstanding). Which is why I believe Teslas are having trouble “seeing”’some obstacles and why pure vision should be better.

 

[magnemoe]

1 hour ago, mikegarrison said:

I drive a manual transmission 2005 Honda Element. It has a radio. That's about it, as far as driver electronics.

A decade to early I have an 10 year Volvo who lack it, my mothers 7 year old has it. 
However google map is much better as it shows real life traffic data on the map.
Still that is not good enough as some sites use the google map and overlay speed traps and common police control points. 
I want the main screens to be android devices, not likely as the car companies can not charge an grand to update them at services. 
Chip trimmed cars is pretty common, in short car companies only make an few engines as they are very expensive to develop, in Europe its very common to tax cars based on their horsepower. 
So they simply nerf the car and sell it cheaper. But you visit an dealer who can do adds and remove this. Remember to use the remote to reset to default before service or EU control. 
Car companies has no interest in making the hack hard as its not main steam so not impact their bottom line and probably get them sales from people whod else would by an used older car. 
 

[mikegarrison]

6 minutes ago, magnemoe said:

A decade to early I have an 10 year Volvo who lack it, my mothers 7 year old has it. 
However google map is much better as it shows real life traffic data on the map.

Well ... I don't have a smartphone either. So I kind of have to do my navigating the old-fashioned way, by looking at signs.

Edited August 31, 2021 by mikegarrison

[magnemoe]

10 minutes ago, StrandedonEarth said:

Re radar vs vision: As I understand radar, it sends out microwaves which reflect back to the sensor. If it hits an angled metal surface (like the top of an overturned semi) most of the radar signal will reflect away from the car sending the signal, so the radar won’t “see” it. Unlike light, which scatters in all directions (mirrored surfaces notwithstanding). Which is why I believe Teslas are having trouble “seeing”’some obstacles and why pure vision should be better.

 

The top of an overturned semi with an container will not give much radar return, however this will be weird as mostly it would have the bottom facing you. 
Main benefit of radar is: large stationary metal stuff on the road=very bad: need to brake now. 
Ran into this once, and its an setting an AI would get problems with. Stationary car just before an road to the left,  some cars towards me, front car not using flasher as low on flasher oil. 
It was during heavy snow so low visibility once I got that the other car was stationary my option was to ditch the car, luckily it was lots of snow between me and the 3 meter to the walkway so it was an soft impact with no damage to car. 
I called for road help then realized I need to put up an warning triangle so others did not rear end me, other car had no warning triangle. 

A shot time after, I called work and told I would be late, seeing an bus run off the road too, the police came and they realized the issue, we tried to push the offending car off the road, then the karren who owned the car arrived with an friend to tow it. Police was to overwork to confiscate her driving licence it was an bus stop 30 meter ahead and back then you could use the starter engine for emergency movement. 

Well I got rescued by the same truck who rescued the bus, he was blocked by an queue so he got me on the road without any effort, it was an 4 axle truck for rescuing trucks an buses  

[CatastrophicFailure]

2 hours ago, StrandedonEarth said:

Re radar vs vision: As I understand radar, it sends out microwaves which reflect back to the sensor. If it hits an angled metal surface (like the top of an overturned semi) most of the radar signal will reflect away from the car sending the signal, so the radar won’t “see” it. Unlike light, which scatters in all directions (mirrored surfaces notwithstanding). Which is why I believe Teslas are having trouble “seeing”’some obstacles and why pure vision should be better.

 

IIRC, this is more or less what happened with the couple of cases of a Tesla going under a semi trailer. The radar is pretty neat, it can bounce waves off the road to see the car ahead of the car in front of you, but that makes it too low to see that trailer. The other scenario I've heard is a trucking coming at you on a cross-street. Radar says it's clear, vision says stop! Which to believe?  Conventional wisdom seems to say the best system combines vision and something else, radar, LIDAR, etc. But Tesla has arguably the best AI people in the world working there, if they say pure vision is best, I can give them the benefit of the doubt. Humans, after all, manage to drive with nothing but vision. Now, get the basics down, then we can talk about something like IR cameras in addition to visible light to see through smoke and fog. I'm wondering if the Hardware 4 mentioned for Cybertruck will begin to incorporate just that.

2 hours ago, tater said:

Watching their new approach on the AI day... it's kind of amazing. I'm one of those who tends to think that actual FSD is a lot closer to an "AI complete" problem than many imagine, and now I'm starting to wonder if narrow AI might literally come along for the ride sooner rather than later (in the Dojo space, not in each car, obviously, not nearly enough computer there).

I think the hardware and software will get to FSD alot sooner than the, er, wetware (ie: regulations). SAE Level 4 should be "relatively" easy, making that step to true Level 5 is gonna take rethinking a whole lot of legalese. Or, by that point, we just let the Tesla Robots figure it out. 

2 hours ago, kerbiloid said:

And that's the difference. You are driving for decades. Like a dolphin, with brain hemispheres sleeping in turns, lol. So, an autopilot is an assistance for you.

While most part of autopilot users are amateurs, which can't be simultaneously alert and sleeping at once.

And yet, with over a million Teslas on the road, these kinds of incidents remain extremely rare. As @tater often says, they're nothing but noise. The same kind of people getting distracted while using Autopilot would probably be distracted and/or otherwise doing dumb things with or without it. The vast majority of people, after all, are quite capable of safely operating an automobile. I would posit that Autopilot makes anyone a better, safer driver: if you're already a good, safe driver, AP just give you that extra layer to make you more so, if "you're" an incompetent, irresponsible moron, Autopilot may make you a better driver, but you're still way below the control.

1 hour ago, magnemoe said:

low on flasher oil. 

I hate it when that happens. 

1 hour ago, magnemoe said:

seeing an bus run off the road too

I hate it when that happens, too... 

[mikegarrison]

16 minutes ago, CatastrophicFailure said:

IIRC, this is more or less what happened with the couple of cases of a Tesla going under a semi trailer. The radar is pretty neat, it can bounce waves off the road to see the car ahead of the car in front of you, but that makes it too low to see that trailer. The other scenario I've heard is a trucking coming at you on a cross-street. Radar says it's clear, vision says stop! Which to believe?  Conventional wisdom seems to say the best system combines vision and something else, radar, LIDAR, etc. But Tesla has arguably the best AI people in the world working there, if they say pure vision is best, I can give them the benefit of the doubt. Humans, after all, manage to drive with nothing but vision. Now, get the basics down, then we can talk about something like IR cameras in addition to visible light to see through smoke and fog. I'm wondering if the Hardware 4 mentioned for Cybertruck will begin to incorporate just that.

Robotic vision has been a notoriously hard problem, in large part because people didn't understand how much of human "vision" is actually signal processing. (Optical illusions take advantage of this by fooling the signal processing, making us think we are seeing something we are not.) Bottom line is that human vision is taking a very small signal (essentially a 2D map of pixels on our retinas) and turning it into a 3D model of the universe around us. That's a HARD problem, one that evolution solved over millions of years. It's not been easy for engineers to do in a few decades.

Edited August 31, 2021 by mikegarrison