Emergency Lane Departure - False Positives

[curry684]

I agree that Tesla's programming seems to assume the driver is not paying attention and/or is bad driver.

Which is of course a quite sane assumption given that 80% of the people on public roads are bad drivers and/or not paying attention, and 80% of the remainder is a bad driver without realizing it, and is quite vocal on online forums about robots telling them they're wrong based on logic and physics

 

[Dogen]

Oh, it wasn't a false positive. It was a calculated and intentional (but very minor) risk. The car just didn't like what I was doing and took away my agency - This is why I'm asking the sort of philosophical question - is it OK for the car take away your agency if it doesn't like what you're doing?

At this stage in the capability of the car to out-think a human? No, it is not ok.

For those who wish to participate in beta testing on public roads, more power to ya. Thanks for your help. Just do no harm.

For those of us who prefer to drive ourselves, (and more specifically me) we have that right, value it very, very highly, and strive to do no harm every time we get behind the wheel. This feature caused my first drivers remorse since picking up the car last November. We need to be able to set the default to “off” and save it to our driver profile. Period. I’ve emailed customer support twice, engaged service twice and will continue to do so till we can.

Imagine this happening to an ICE luddite on a test drive..... or a reporter for Business Insider, CNBC, et cetera.

 

[Msjulie]

Which is of course a quite sane assumption given that 80% of the people on public roads are bad drivers and/or not paying attention, and 80% of the remainder is a bad driver without realizing it, and is quite vocal on online forums about robots telling them they're wrong based on logic and physics

That wasn't nearly as clever as you thought it was.

 

[diamond.g]

As this is a safety issue, has anyone contacted Consumer Reports to see if they can post something about it?

 

[ammulder]

I have been afraid to ask this question. I always use my signals...I've never had an ELDA activation. Coincidence?

I’m on .16.3

My wife has complained about ELDA activating for her. It’s never done it for me. Sure enough, I was riding along with her, and it triggered on a spot I drive by all the time. She said it’s done it at more or less the same spot before for her. This is on a 1-lane-each-way reasonably busy rural road — good stripes but no shoulder and detectable things right beside the road (foliage, posts for electric lines, etc.). There was no lane change, no danger — a real false positive.

So I feel like it’s more a matter of driving style than blinker use. I don’t know what it is about her that sets it off, but I’ll try to remember to save the footage if I’m along for it again.

Really would like .16.3.2 or .20.1 since people have said they’re better!

 

[smogne41]

As this is a safety issue, has anyone contacted Consumer Reports to see if they can post something about it?

I would rather give Tesla a little while to make it right first. We have a work around for now (turn off every drive). Will consider escalating if they double down on this though.

 

[curry684]

Haven't had a single false positive since 2019.20 anyway.

 

[jyalpert]

Which is of course a quite sane assumption given that 80% of the people on public roads are bad drivers and/or not paying attention, and 80% of the remainder is a bad driver without realizing it, and is quite vocal on online forums about robots telling them they're wrong based on logic and physics

I just don't know about this. It's difficult to say what a "good" or "bad" driver is, unless you were to, say, document a long list of behaviors that have X probability for an accident, weighted by severity, and then follow every driver around for a statistically significant amount of time, in order to evaluate if they are better or worse than average. Then, you'd have to evaluate whether or not Tesla's driver aids were making improvements on those activities, and to what extent.

In other words, it's not really measurable in any practical sense. Hence the moral conundrum. We all want self-driving cars to be better than humans at driving. But it's not really possible yet to say if these driver aids are a step in the right direction or not.

Assuming we're all a**holes is fine, but without any information to say whether the technology is making us better, we will just end up arguing. Really your perspective just reflects whether your null hypothesis is "any new driver's aid is better" or "any new driver's aid is likely worse".

Like, I've been driving for 25 years, probably hundreds of thousands of miles in many environments, on several continents, never been in an accident, never gotten a speeding ticket or other moving violation, fairly competent advanced-group track day driver, done a bit of wheel-to-wheel racing, etc. But none of that proves I'm a good driver or a bad driver. Maybe I've just been lucky and attribute it to skill. Maybe some of these aids would actually make me better. Or not. But nobody can say at this point. So all we have is this - arguments about our ideas and feelings about the technology.

 

[CarMAN3]

Which is of course a quite sane assumption given that 80% of the people on public roads are bad drivers and/or not paying attention, and 80% of the remainder is a bad driver without realizing it, and is quite vocal on online forums about robots telling them they're wrong based on logic and physics

Hahahahaha

 

[jyalpert]

I saved it to the SD card afterward, I'll see if I can pull it.

Alright here we go.

#1 - ELDA tries to force me to abandon a lane change:

#2 - ELDA interferes when I'm taking a wide line on an on-ramp:

#1 was dangerous, #2 was annoying.

 

[SDKoala]

I haven't been following this thread closely and only saw it exists just now. I feel like I've found my people!

I wrote this in the Firmware 20.2.1 thread:

I left ELDA on yesterday to see if there was any improvement in the false positives. There are not. It still gets confused in the same places on my commute.

There is a difference, though. It didn't jerk the car forcefully back in the other direction to "correct" my driving. As far as I could tell, it didn't move the car at all as I didn't feel any pull in the opposite direction like before. It still sounded the alarm though. So basically, for me, it's gone from a dangerous feature that has a chance of putting me too close to traffic traveling in the opposite direction to a useless feature that can't be trusted to do what it is supposed to do if there actually is a real hazard to avoid.

 

[MP3Mike]

Alright here we go.

#1 - ELDA tries to force me to abandon a lane change:

As far as I know that was an illegal lane change, so I can sort of see why it would do that. It may be thinking that you are overshooting your lane change so it was trying to keep you in the lane you were changing to.

My understanding is that in most states you can only legally change lanes one at a time and you are supposed to cancel your turn signal, wait at least 100 feet, and then signal again before the next lane change.

 

[jyalpert]

As far as I know that was an illegal lane change, so I can sort of see why it would do that. It may be thinking that you are overshooting your lane change so it was trying to keep you in the lane you were changing to.

My understanding is that in most states you can only legally change lanes one at a time and you are supposed to cancel your turn signal, wait at least 100 feet, and then signal again before the next lane change.

Sure. It wasn't the best or safest lane change. (I address this in a previous post in this thread).

That's not really at issue here. The issue is that ELDA did not like my lane change and made the entire thing MORE dangerous by trying to force me to abandon it after I'd already committed. You can see how confused the guy behind me becomes.

I would love to drive in an ideal manner to the letter of the law all the time, but I don't. So if they built a system where, if you don't obey 100% the letter of the law, it makes it more dangerous, is that OK?

 

[Big Earl]

Alright here we go.

#1 - ELDA tries to force me to abandon a lane change:

#2 - ELDA interferes when I'm taking a wide line on an on-ramp:

#1 was dangerous, #2 was annoying.

#1 you took two lanes at once while there was a vehicle approaching at a higher velocity in the second lane. Its correction did not appear to put you in any sort of additional danger.

#2 you cross the line and it responded appropriately.

 

[jyalpert]

#1 you took two lanes at once while there was a vehicle approaching at a higher velocity in the second lane. Its correction did not appear to put you in any sort of additional danger.

#2 you cross the line and it responded appropriately.

Welcome to the discussion. We already know it has real false positives. This is not that. It's working as intended in these situations. The question is whether the intent is reasonable - that any time it is triggered, the assumption is that the driver is not acting intentionally. Is that OK?

Also I would argue that it did put me in additional danger in situation #1. The guy behind me clearly speeds up and moves to the edge of the lane to overtake when I shift back over to the right.

 

[Economite]

It's working as intended in these situations.

I would argue that it wasn't "working as intended" in your situation. The purpose of this feature is to do corrections when a driver drifts into another lane, not to interfere with intentional steering choices that cross the lane line. You were intending to cross the line and change into the second lane; but the car couldn't figure that out. That makes this a kind of false positive. The system isn't smart enough to be able to do what it is intended to do, so it is overriding intentional maneuvers.

 

[jyalpert]

I would argue that it wasn't "working as intended" in your situation. The purpose of this feature is to do corrections when a driver drifts into another lane, not to interfere with intentional steering choices that cross the lane line. You were intending to cross the line and change into the second lane; but the car couldn't figure that out. That makes this a kind of false positive. The system isn't smart enough to be able to do what it is intended to do, so it is overriding intentional maneuvers.

Yes, this is the core of it. It's working as intended from a superficial perspective, in that it is identifying a situation and reacting to that situation. The issue is that there is no understanding of driver intent.

But does that matter? For a safety system like ABS, it's easy - stopping in a shorter distance is always better than a long distance. For TCS, it's also pretty easy - some drivers might want to exceed the limit of grip of the tires, but in almost all cases on public roads, that is a dangerous situation. For ELDA, it's not obvious - you could be doing relatively normal driving, and the system can take away your agency.

Arguments I could see in favor of this "don't care about intent" beahvior:

• It will teach bad drivers (apparently me) to drive better
• If this is the only way to get the capability into the hands of truly poor drivers, other drivers who are bothered by it are just collateral damage

The philosophical issue for me arises if the system can take a situation and make it MORE dangerous. Which is why situation #1 is concerning to me. Before that, it was all situation #2 or freaking out at parked cars. But with situation #1, you might be in a trolley-problem type of situation.

I think I would be more comfortable with the situation if Tesla had some really convincing data about the system + formal training. Like - X# of severe accidents with Y typical outcome occur due to unaware or distracted drivers drifting from their lane. ELDA is able to prevent Z% of these events, although it has the potential to cause an additional Q% of accidents when used improperly. And then they would provide required online training for all driver profiles, explaining situations where the system will actually cause danger, and then you'd consent to the use. Instead this thing was just uploaded to my car in the latest software release with no explanation.

 

[Sophias_dad]

I've now had a chance to repeat (by accident) an ELD false positive, on the same road. On a gently twisty but relatively narrow road with very clear fog lines and a long line-of-sight down the road, I apparently vary my distance from fog line too much for ELD to tolerate(as the road turns right and then left a foot or three, and it thinks I'm gonna go careening off the edge. It might also be crack-seal stripes in the lane, but I doubt that a bit.

 

[hcdavis3]

I haven't been following this thread closely and only saw it exists just now. I feel like I've found my people!

I wrote this in the Firmware 20.2.1 thread:

I left ELDA on yesterday to see if there was any improvement in the false positives. There are not. It still gets confused in the same places on my commute.

There is a difference, though. It didn't jerk the car forcefully back in the other direction to "correct" my driving. As far as I could tell, it didn't move the car at all as I didn't feel any pull in the opposite direction like before. It still sounded the alarm though. So basically, for me, it's gone from a dangerous feature that has a chance of putting me too close to traffic traveling in the opposite direction to a useless feature that can't be trusted to do what it is supposed to do if there actually is a real hazard to avoid.

So far this is the most stable yet for me.AP works better. Homelink is better. Backup Cam works better. So far so good.

 

[SDKoala]

So far this is the most stable yet for me.AP works better. Homelink is better. Backup Cam works better. So far so good.

That's nice, but what does it have to do with the subject of the thread?