New cars can stay in their lane—but might not stop for parked cars
- Thread starter JournalBot
- Start date
*Checks for Beth Mole as author.*
On that though, maybe the number of Tesla accidents/fatalities could also be to do with Tesla overselling the autopilot functionality?
Upvote
57
(82
/
-25)
I wish they had included tests with Toyota's ADAS implementation in this roundup, curious where it would land. In my experience with an '18 Highlander (very sensitive to stationary - eg stopping to turn - objects) and a '20 RAV4 (not quite as sensitive but with auto breaking to "hold" at a stop light with car stopped in front) I've been impressed with the functionality.
Lane keeping on the other hand... Super annoying and I just turn it off.
At the end of the day this tech is helpful but if one relies on it to do the job your eyeballs and brain should be doing while moving quickly in a multi-thousand Kg object then you're just asking for trouble.
Edited for readability. Not enough
️
Lane keeping on the other hand... Super annoying and I just turn it off.
At the end of the day this tech is helpful but if one relies on it to do the job your eyeballs and brain should be doing while moving quickly in a multi-thousand Kg object then you're just asking for trouble.
Edited for readability. Not enough
️
Upvote
17
(18
/
-1)
I've got a 2019 BEV with the full suite of driver assist. I absolutely love that stuff for long highway drives, as it reduces driver fatigue. But I have learned to play close attention when coming up on a traffic jam and have my foot on the brake pedal, as the car sometimes seems slow to react to slowed/stopped traffic in front me. That's been the case with every car I've driven with that tech.
You must always remain vigilant.
You must always remain vigilant.
Upvote
60
(62
/
-2)
Seems like these primitive systems are inconvenient and somewhat dangerous.
But, what I want to really know is, what does this mean precisely and exactly:
"...cars from GM and Ford had geofences that prevented the ADAS from engaging..."?
But, what I want to really know is, what does this mean precisely and exactly:
"...cars from GM and Ford had geofences that prevented the ADAS from engaging..."?
Upvote
7
(14
/
-7)
Hey, it's a lane-keeper, not a don't-crash-into-objects-right-in-front-of-you-er! Wile. E. Coyote does not see the problem.
/s
"Best system went two out of three" is remarkably damning. Software engineers from a fast-and-loose discipline playing with safety-of-life equipment with little apparent appreciation of the gravity of their product. Most concerning that it seems like basically every manufacture is drawing engineers from this talent pool without effectively reigning them in.
/s
"Best system went two out of three" is remarkably damning. Software engineers from a fast-and-loose discipline playing with safety-of-life equipment with little apparent appreciation of the gravity of their product. Most concerning that it seems like basically every manufacture is drawing engineers from this talent pool without effectively reigning them in.
Upvote
27
(45
/
-18)
I wonder how much of the problem is unrealistic expectations. I really like the adaptive cruise control on my '17 Accord. But I also went into the purchase fully aware that was I was buying was just a marginally less stupid form of the cruise control in the cars i learned to drive in 20-25 years ago. It lets me not have to constantly feather the brake/gas in light to moderate traffic to compensate for the person I'm following not being able to maintain a constant speed. I was never under the delusion that it would let me turn my brain off.
Upvote
88
(89
/
-1)
It increasingly looks like self-driving cars are an all-or-nothing proposition, not necessarily due to technology, but due to the human factors.
I think that's a lesson that manufacturers will be very reluctant to learn, so I hope that regulators are up to the job of drawing a live between low-level driver assists like conventional cruise control that require drivers to remain engaged and more extensive assists that tempt the driver to let their attention wander.
Along with the widely reported Tesla incidents, the idea of cars plowing in to a car stopped on the side of the road would be terrifying if I had to stop and change a flat.
I'm not sure where that line should be. I do know that when I've driven a car with adaptive cruise control and lanekeeping, I've definitely felt less involved moment-to-moment, but I honestly don't know if my attention wandered to a dangerous degree.
The more I learn about this area, the less inclined I am to want additional ADAS features in my next car or in the cars around me.
I think that's a lesson that manufacturers will be very reluctant to learn, so I hope that regulators are up to the job of drawing a live between low-level driver assists like conventional cruise control that require drivers to remain engaged and more extensive assists that tempt the driver to let their attention wander.
Along with the widely reported Tesla incidents, the idea of cars plowing in to a car stopped on the side of the road would be terrifying if I had to stop and change a flat.
I'm not sure where that line should be. I do know that when I've driven a car with adaptive cruise control and lanekeeping, I've definitely felt less involved moment-to-moment, but I honestly don't know if my attention wandered to a dangerous degree.
The more I learn about this area, the less inclined I am to want additional ADAS features in my next car or in the cars around me.
Upvote
39
(48
/
-9)
Looks like self-driving is going to be another perpetually 10 years away technology. Good to see this kind of all too rare third party testing bringing a dose of reality.
I accept the tech is genuinely advancing but it’s going to take longer than people think. Microchips are around 60 years old and electronic digital programmable computers are about 80 years old at this point. It could be 30 years before cars are fully autonomous.
FYI self-driving cars were trialled in 1957. They followed a cable embedded in the highway, and were able to detect stopped cars in front of them. But exactly as in this Ars article, it didn’t always work properly ...
https://spectrum.ieee.org/tech-history/ ... nerin-1960
I accept the tech is genuinely advancing but it’s going to take longer than people think. Microchips are around 60 years old and electronic digital programmable computers are about 80 years old at this point. It could be 30 years before cars are fully autonomous.
FYI self-driving cars were trialled in 1957. They followed a cable embedded in the highway, and were able to detect stopped cars in front of them. But exactly as in this Ars article, it didn’t always work properly ...
https://spectrum.ieee.org/tech-history/ ... nerin-1960
Upvote
45
(50
/
-5)
It's clear that all lane-keeping systems have this problem, but Tesla's has led to more crashes.
I'd guess this is down to Tesla's marketing. They talk up their Autopilot, and people buy their cars expecting to use it in that way.
If you're primarily driving the car yourself, and as a bonus it's watching the lane / ready to apply emergency braking, this is a 100% sensible way for the software to behave. If it misses a car in the road, the driver should still see it.
If the car's primarily on Autopilot, and as a bonus you have an eye on the road while your mind wanders, it's more likely both you and the car will miss an obstacle... especially if it's low or only slightly blocking the lane.
I'd guess this is down to Tesla's marketing. They talk up their Autopilot, and people buy their cars expecting to use it in that way.
If you're primarily driving the car yourself, and as a bonus it's watching the lane / ready to apply emergency braking, this is a 100% sensible way for the software to behave. If it misses a car in the road, the driver should still see it.
If the car's primarily on Autopilot, and as a bonus you have an eye on the road while your mind wanders, it's more likely both you and the car will miss an obstacle... especially if it's low or only slightly blocking the lane.
Upvote
2
(42
/
-40)
Most auto-pilots have been a huge let down unfortunately. It should be very easy to design an auto-pilot that would work primarily in traffic jams, which is when most people would want auto-pilot. But instead that seems to be when they're the weakest. I don't need auto-pilot for cruising down country roads on Sunday.
Upvote
-9
(15
/
-24)
There is an old programmer's saying "the first 80 percent of the problem takes 80% of development time; the last 20% takes the real 80% of development time".
Again and again when you think you are "so close" to the end, you find out that it's where the work really begins.
It seems like a bunch of companies solved the "easy 80%" of the problem space (perhaps "easy 90" or even "99%"). Unfortunately, it's very far from good enough, and getting the rest of it is _hard_.
Again and again when you think you are "so close" to the end, you find out that it's where the work really begins.
It seems like a bunch of companies solved the "easy 80%" of the problem space (perhaps "easy 90" or even "99%"). Unfortunately, it's very far from good enough, and getting the rest of it is _hard_.
Upvote
55
(59
/
-4)
Level 3 and 4 systems are MORE dangerous than nothing at all. A system which is good enough for the driver to take their attention off the road, but then demand they pay attention "just in case" is completely ignoring human psychology and asking for accidents. If we can turn our attention away.. WE WILL. In some cases, expecting the driver to take over in an emergency is not possible. The incident comes upon the driver too fast for them to re-engage their full attention, assess the situation, realize the car isn't doing what its supposed to be doing, and react accordingly.
Upvote
86
(102
/
-16)
Seems like if it can't handle those possible cases it's more dangerous than just making the person pay attention and manipulate the controls.
If it works "most of the time" people will become disconnected from the environment and be unable to realize when they are entering a situation they need to take control back or what to do. It's the same problem when driving too long on a straight stretch of empty highway, you get highway-hypnosis if you aren't having to interact with the environment by executing curves, changing lanes, etc.
This seems like a really bad setup when you come up on a broken down or pulled over car and it doesn't notice, people will end up seriously hurt when they plow into them. Or in an urban environment where lanes are sometimes parking, sometimes driving by time of day. Or urban delivery vehicles that park briefly in the driving lane as is so common.
EDIT: When I say "pulled over" I mean like so many I see that are pulled over but sticking out a bit into the lane, or stopped by a cop and the cop positions their car at an angle part way into the lane to deflect anyone that may hit him walking up to the stopped car. Frequently the lane is still partially or completely obstructed depending on the situation.
If it works "most of the time" people will become disconnected from the environment and be unable to realize when they are entering a situation they need to take control back or what to do. It's the same problem when driving too long on a straight stretch of empty highway, you get highway-hypnosis if you aren't having to interact with the environment by executing curves, changing lanes, etc.
This seems like a really bad setup when you come up on a broken down or pulled over car and it doesn't notice, people will end up seriously hurt when they plow into them. Or in an urban environment where lanes are sometimes parking, sometimes driving by time of day. Or urban delivery vehicles that park briefly in the driving lane as is so common.
EDIT: When I say "pulled over" I mean like so many I see that are pulled over but sticking out a bit into the lane, or stopped by a cop and the cop positions their car at an angle part way into the lane to deflect anyone that may hit him walking up to the stopped car. Frequently the lane is still partially or completely obstructed depending on the situation.
Upvote
22
(29
/
-7)
I drive a 2020 Ford Escape hybrid with lane keeping and adaptive cruise, and this mirrors my experience completely. The lane keeping system is both weirdly conservative - needing too much pressure on the wheel to just loosely hold it (like you would on long highway stretches) - but also has a tendency to make mistakes if road lines disappear unexpectedly or if the shoulder tapers off to gravel. Adaptive cruise has mostly been fine, but the one time I let it stop the car at a light was exactly as nerve-wracking as the article mentions.
Maybe being unreliable is a safety feature, since I never trust any of those systems enough to really disengage mentally from driving.
All of that said though, I still use those features all the time for highway driving and wouldn't buy a car without them.
Maybe being unreliable is a safety feature, since I never trust any of those systems enough to really disengage mentally from driving.
All of that said though, I still use those features all the time for highway driving and wouldn't buy a car without them.
Upvote
20
(25
/
-5)
Cadillac/GM has pre-mapped 200,000 miles of roads of highways around the US. Super Cruise only engages on these pre-mapped roads. This improves safety because it makes it less likely the system will get confused by a new situation it hasn't encountered before. I know less about the Ford system but I assume the situation is similar there.
Upvote
52
(53
/
-1)
I've got a new vehicle on the way, and strangely enough that on the way to the dealer the wife got stuck behind someone who kept speeding up and slowing down. The wife was so annoyed because she had to keep adjusting the cruise control.
I will be using this example to teacher her to treat this as "adaptive cruise control" and not some sort of self-driving system. The dealer did describe it this way, so I'm glad for that.
I will be using this example to teacher her to treat this as "adaptive cruise control" and not some sort of self-driving system. The dealer did describe it this way, so I'm glad for that.
Upvote
7
(9
/
-2)
FYI, that kind of detail would be a welcome sidebar or footnote in the story.
Either way, though: thanks for clarifying! I learned something
Upvote
13
(16
/
-3)
The GM system uses a whitelist model, and only operates on highways that they've pre-mapped.
Upvote
44
(44
/
0)
I own a 2018 Volkswagen Tiguan that includes lane-keeping assistance and adaptive cruise control; together they're pretty clever for keeping me aligned and crash-free on US highways. However, it's not just "parked cars" this combination of technology won't see. If you're using this tech on surface streets and come upon a car stopped at a red light that the vehicle didn't prior see moving, it simply doesn't see the vehicle at all. Without driver intervention, it would simply plow through the stopped vehicle. I've become accustomed to this behavior and adjust for it and lane-keeping + adaptive cruise control are NOT fully-autonomous driving. I briefly mentioned this phenomenon to Volkswagen and was told if I wanted fully-autonomous, I should buy a Tesla.
There is no substitute for paying attention while driving, even if your car is performing most of the tasks autonomously. Anyone who crashes into a parked car or a stopped vehicle is culpable for the results.
There is no substitute for paying attention while driving, even if your car is performing most of the tasks autonomously. Anyone who crashes into a parked car or a stopped vehicle is culpable for the results.
Upvote
37
(37
/
0)
You're legally required to monitor and intervene. It's your job as a driver. Next we'll be seeing athletes complaining about the need for frequent exercise and training.
Upvote
3
(31
/
-28)
I just got a 2020 Escape Hybrid and had exactly same experience. I did make one attempt to let car stop on its own at red light, with a car in front of it. By the time I would have hit brakes myself, it kept going full speed, then slammed on the brakes, about the time I would have slammed on them myself. It stopped just fine, but less efficient and more annoying to other drivers.
I think my plan from now on is to just use adaptive CC and not bother with lane-keeping, when on very long drives. This will force me to keep engaged with the road, without having to micro-manage the throttle.
Self-Park, on the other hand, was flawless, when I tried it. Did far better than I would have, and much faster too.
Upvote
24
(24
/
0)
I know I'm a just bit of a thicko in the cheap seats, but I'm glad they've only enabled it on mapped roads, because it seems to me nothing ever changes in real life.
Upvote
8
(13
/
-5)
It's not only false negatives which would worry me with the current generation systems, false positves can be just as dangerous.
Like here: https://youtu.be/vGLehCMiZy4?t=40
Like here: https://youtu.be/vGLehCMiZy4?t=40
Upvote
5
(5
/
0)
Uh oh, I think the AI has figured out the cars were fake the whole time …
just like the exocomps!
just like the exocomps!
Upvote
-8
(2
/
-10)
If drivers are complaining about the need to actually pay attention, maybe manufacturers are being dishonest in their disclosure of what their vehicles are actually capable of. Looking at you, autopilot.
Or people are just stupid.
Or people are just stupid.
Upvote
14
(25
/
-11)
Do you mean Level 2 and 3?
My interpretation is that Level 4 will end up being the safest actually attainable level we can get, since it will be possible to operate autonomous vehicles within geofenced areas that keep them on safe roads and operating in weather conditions that are amenable to their sensing equipment. By adding the significant (and fairly arbitary) constraint of only functioning within their "operational design domain (ODD)," it seems like Level 4 could end up being a very robust option that will actually happen in a decade or two. This is realistically what all the major players (Waymo, Cruise) are pursuing for their commercial robo-taxi plans. More than a decade of research and billions upon billions of dollars spent have drastically chastened their ambitions as they grew to understand the limitations of the systems we have at our disposal.
I am not sure that anyone involved in producing these systems is truly confident that a Level 5 system - one that can operate in any geography, in all conditions, with no intervention - is actually feasible in the foreseeable future (call it 20 years out). Elon Musk's bluster in this area is mainly for the sake of publicity.
Upvote
18
(22
/
-4)
That's not a conclusion that can be drawn at all about Tesla's system. It's extremely doubtful that other brands of cars crashing, in exactly the same circumstances as the Tesla crashes that have received extreme scrutiny, would be investigated to the extent the Tesla crashes have been. Random non-Tesla sedan hitting a semi crossing the road - "oh, another driver wasn't paying attention, oh well, very sad. Move on". Random non-Tesla sedan hitting a highway exit/lane splitting wall - "oh, the driver wasn't paying attention, oh well, very sad. Move on." No one will assume they had an ADAS system engaged in other brands of vehicles because it's not standard equipment like it is on a Tesla. Other brands aren't anywhere near as conspicuous due to novelty in the market either.
The question to ask of authorities like the NTSB is, in the case of accidents where the apparent cause is drive inattentiveness, how many are investigated deeply enough to learn if an ADAS system was engaged prior to the accident? Are logs from the cars computer checked as a matter of course in all such accidents? Do other car brands even keep logs of ADAS usage that are retrievable?
Upvote
19
(31
/
-12)
I don't think it's that, exactly. I think it's more that they've evolved their technology -- all these companies -- out of systems never intended for self driving, and they're doing the least important parts first - cruise control, then adaptive cruise control, then add lane keeping and so on, and navigation. Those are all nice to have, but the number one priority is detection and collision avoidance with people and cars. Zero hours should have been spent on the other features until they had that to "works better than a human paying close attention." Then you give that system absolute priority over other systems as you develop them and add them into the system.
It's as if these systems are being developed by people who never drove a car themselves, but I know it's not really that. They're doing the easy stuff first because it's easy, not because it's important.
Upvote
14
(21
/
-7)
I'll let you guys do the beta testing of these features and in 20 years, when the tech is actually safe, I might use it.
Upvote
8
(13
/
-5)
I don't really understand the point. If you're required to continue paying attention all the time...what benefit is there? You don't have to adjust the wheel or the speed 95% of the time, sure....but you don't know when that 5% will show up so you're still mentally tuned in.
Seems like a lot of technology / expense for very little reward.
Upvote
19
(27
/
-8)
Upvote
-13
(6
/
-19)
We have a 2019 Toyota Highlander with these features (adaptive cruise and lane departure). The adaptive cruise control is fine, as it mainly deals with the speeds of other vehicles. The lane departure is terrible, however, and we've turned it off. It gave way too many false warnings and alerts, particularly when highway striping is out of alignment (construction zones in particular, but also when the line has been painted over and re-positioned). It proved pretty much useless and only made driving more stressful. I would rather just pay attention than have an unreliable system try to catch my mistakes.
If a future car has this feature, I won't buy it if I can't turn it off.
Upvote
21
(22
/
-1)
ColdWetDog
Ars Legatus Legionis
How often do they recheck the roads? Seems like DOTs everywhere just love to improve things. Often temporarily.
Upvote
11
(12
/
-1)
Yeah. This is what makes me laugh at just about every car commercial these days. They show a car drifting out of a lane and wow the car corrects itself! So cool! Or they show a car driving through a parking lot and someone steps in front of it, and the car stops itself! Amazing!!
I find both these scenarios to be ridiculous. Both are showing inattentive drivers who should have their licenses suspended. Staying within a lane and not running over people should be a minimum requirement for driving a vehicle, not something you rely on a half-baked automated system to help you with!
I'm all for fully self driving cars, but I'm pretty much all against all these halfway measures that are being used as selling points these days.
Upvote
40
(45
/
-5)
Except the data doesn't show this at all. If people were taken in by the marketing, we would see a clustering of malfunctions with new drivers or people who borrow the car to test out the thing with the cool name. In reality, people who use the system for the first time tend to be very careful and most drivers seem to work their way up, getting more and more comfortable over time. At the same time, all the reported incidents were of people who were definitely aware of the limitations and chose to ignore them (often while engaging in other behavior that is definitely illegal/ill-advised regardless of what you car is capable of).
It makes for a nice story, but when every investigated case is someone who is experienced with the system (sometimes going so far as to participate in bug reports) and shows a blatant disregard for safety (outside of just their use of Autopilot), they're just as likely to cause accidents with a different system or none at all. If it was something that impacted people who clearly only know it from the marketing (like the tales of people crashing their new Porsche on the first corner), I'd buy it. But so far, we haven't seen an instance of grandma or a beginner driver pressing the Autopilot button the first time out and expecting the car to drive itself. As it stands, the biggest marketing misconception in regards to Autopilot might be that it attracts lawyers who offer to throw in the name to get you off and, so far, have always lost when the actual data came in.
Upvote
16
(24
/
-8)
Driving is the most dangerous thing most of us do regularly. Self-driving cars is one solution.
A better solution: Have rigorous standards for driving licensure. It currently takes zero skill to get a license, so lots of people die.
This is a case of tech and car companies spending billions to try to fix a problem that's entirely government-created, and easily solved.
A better solution: Have rigorous standards for driving licensure. It currently takes zero skill to get a license, so lots of people die.
This is a case of tech and car companies spending billions to try to fix a problem that's entirely government-created, and easily solved.
Upvote
7
(22
/
-15)
In 2010 I believed self-driving cars were going to revolutionize driving. They would be safer than human drivers and that meant insurance companies would charge humans who wanted to drive a premium. This would lead to a generation where most humans would have no idea how to drive a car. And I believed this would come to pass by 2030 at the rate of learning.
Now it appears these technologies have actually made driving less safe because they're not fully developed yet, but are being included in production models and marketed to the public. These features seem to give at least a percentage of drivers a false sense of security, thus making the roads less safe.
And I now think it will be several decades before we're to the point where the digital driver is safer than the human drive.
Now it appears these technologies have actually made driving less safe because they're not fully developed yet, but are being included in production models and marketed to the public. These features seem to give at least a percentage of drivers a false sense of security, thus making the roads less safe.
And I now think it will be several decades before we're to the point where the digital driver is safer than the human drive.
Upvote
2
(8
/
-6)