7 ms·

I'm currently driving a pretty new Mercedes (company car), and its lane assistant regularly tries to kill me in construction areas because it wants me to follow

I'm currently driving a pretty new Mercedes (company car), and its lane assistant regularly tries to kill me in construction areas because it wants me to follow the regular, currently inactive lane markings. It's gotten so bad that I'm switching off the lane assistant when I'm approaching construction areas. I've talked to a guy at the car dealership, and he said "my mother complains about the same thing, it's a known issue". I really, really hope they've managed to fix that for the autonomous driving models, or at least auto-disengage before construction areas. (And if they fixed it, can I get a software update please?)

theta_d 5y ago
My 2021 Honda Accord does the same. Also turned it off.
- LightG 5y ago
  And ... this is why I've always said it's either all or nothing and why FSD and similar are pointless until in, who knows, 10 years time(if that) it becomes ready for prime time. Sure, youtubers can video it for fun and drink the kool aid but actual drivers will just turn it off after a hairy situation. Well, I will anyway. Maybe this is why Tesla are touting their ridiculous robot ... it's a potential pivot or alternative use of the (no doubt) impressive AI tech behind FSD which, they've now realised, will take much longer to be properly driver ready. Maybe FSD's fate is similar to the space program. The eventual secondary technology it spawns is more useful than its original purpose. Because I won't be going near these autonomous driving solutions for many years. Yeah, give me a Dalek-style home robot instead for now.
ajross 5y ago
As far as I can tell, this Drive Pilot thing is precisely the existing lane assist product. What's changed is that they've heavily constrained it so that it disengages if you go over 60 kph, or outside the hand-maintained list of blessed areas. Presumably they feel they can detect and disallow construction areas and other complexities fast enough to avoid accidents. Which is to say... this is mostly a stunt. Teslas are literally driving people around now[1], and the rest of the industry feels they need to do something. Announcing a "SAE Level 3" product, no matter how constrained, at least gets them marketing hits like this that look like an advantage. [1] No joke: mine takes my kids to and from school reliably. FSD beta isn't finished, but it's really, really good. There remain some path planning and confidence glitches that force me to disengage every dozen miles or so, but in 500 miles since I got it I've yet to see the car attempt anything genuinely unsafe. Mostly it just annoys other drivers by refusing to enter traffic.
- resoluteteeth 5y ago
  If Mercedes-Benz is able to define conditions where their self driving software is guaranteed to work fully autonomously without human intervention and Tesla can't, that in itself is an actual advantage over Tesla's system even if Tesla's system can do more with human monitoring in other conditions. If Tesla is also able to get approval for this (which might be hard if they push OTA updates that completely change the way the software works all the time and can't guarantee there won't be regressions) then they should by all means do so.
  - ajross 5y ago
    That's sort of true in the abstract, but misses the point. Autonomy isn't about "defining conditions", it's about solving real problems. Cars that are L3 only in traffic jams on the Autobahn aren't very autonomous, nor useful. If it's stop go and the system gets above 60kph, it cuts out. If the stoppage is due to construction, it won't work. I mean, sure, it's an advantage for people who actually need to drive in exactly those conditions. But it's not a engineering advantage of the system. Fast forward a year and look at where Daimler will be and where Tesla will be. Who's going to get to a L3 system on general roads first? Who's going to be running L4 driverless vehicles first? Which is to repeat: it's just a stunt. It's a way to "sound like" they're "ahead" in this area of technology, when clearly they aren't. In fact this car is doing more or less exactly what Teslas were doing in the first version of Autopilot more than six years ago. And note that in all that time, those Teslas haven't been hitting anyone in slow traffic jams on well-maintained limited access highways either.
    - resoluteteeth 5y ago
      > That's sort of true in the abstract, but misses the point. Autonomy isn't about "defining conditions", it's about solving real problems. Cars that are L3 only in traffic jams on the Autobahn aren't very autonomous, nor useful. If it's stop go and the system gets above 60kph, it cuts out. If the stoppage is due to construction, it won't work. > I mean, sure, it's an advantage for people who actually need to drive in exactly those conditions. > But it's not a engineering advantage of the system. Fast forward a year and look at where Daimler will be and where Tesla will be. Who's going to get to a L3 system on general roads first? Who's going to be running L4 driverless vehicles first? > Which is to repeat: it's just a stunt. It's a way to "sound like" they're "ahead" in this area of technology, when clearly they aren't. In fact this car is doing more or less exactly what Teslas were doing in the first version of Autopilot more than six years ago. And note that in all that time, those Teslas haven't been hitting anyone in slow traffic jams on well-maintained limited access highways either. If it's a race to see which company can develop fully self-driving cars that work in all conditions in the next 5 years than maybe Tesla could be "ahead" of Mercedes-Benz (although it seems to be way behind other companies like Waymo). However, if we are actually decades away from fully self-driving cars that work in all conditions it is much more useful to have cars that can be trusted to operate without human supervision in limited but well defined conditions, because it is much more useful to be able to have a car that allows you to do something else other than focus on driving SOME of the time than a car that assists you more in some ways but requires supervision ALL of the time. If Tesla's approach precludes them from making self driving technology work reliably under limited conditions like this, it doesn't matter whether their self driving technology works better (but not reliably) in other conditions, because in practice it won't be as useful as more limited but reliable self driving systems. Some countries are already requiring the approach Mercedes-Benz is taking here for regulatory approval and based on the number of accidents Tesla vehicles have been involved in its possible that the US could adopt the same approach. If this happens it will not be useful at all to have a more advanced but incomplete and unreliable self-driving technology in the short term (but perhaps it will be possible to keep developing it without shipping it in cars until it reaches a sufficient level of reliability).
globular-toast 5y ago
It's more likely to kill a worker than you.
nradov 5y ago
Report it as a safety defect to the NHTSA (or your national regulator depending where you are). If enough people file similar reports then they do investigate and require manufacturers to recall the vehicles. https://www.nhtsa.gov/report-a-safety-problem https://www.nhtsa.gov/report-a-safety-problem
chollida1 5y ago
So does my Tesla:( I always take control when entering any construction site. I don't think any other than waymo has solved this yet. My telsa is good but the amount of times I have to take control from it is way too high. Snow is another area where autonomous driving sucks at the moment, and as a Canadian, man that makes my Tesla's self driving not useful 5 months of the year. Waymo seems to be way out front and everyone else seem to be fighting for second here. Though I'm very happy that we're atleast at this stage and not where we were 10 years ago. I just view it as we now have working lane keeping and adaptive cruise control for most new cars these days and that's a solid win.
- ra7 5y ago
  > Waymo seems to be way out front and everyone else seem to be fighting for second here. I'd say Cruise is comfortably second. They've made very good progress in SF and I believe they are doing driverless rides there for employees.
  - rconti 5y ago
    That's correct, Cruise is doing driverless rides for employees in SF. I didn't know about Waymo's driverless rides, though-- can't say I follow the space that closely. Wow!
  - sieabahlpark 5y ago
    Isn't that the company that stole IP from waymo?
- cycrutchfield 5y ago
  Just curious, what is your personal experience with Waymo that leads you to this conclusion? Also curious to know about Waymo’s performance in the snow because I haven’t heard much about that.
  - nopzor 5y ago
    just took a waymo fully autonomous (no safety driver, nobody in the driver seat, no manual override) taxi in scottsdale to lunch. was surreal. supposedly over 1m miles with no faults.
ct520 5y ago
My new ram limited does the same thing
a4isms 5y ago
I use “Pilot Assist” in a 2018 Volvo V90CC, and on the one hand, it has all kinds of “issues” that require my active attention, including confusion about construction areas, or even just a general wandering when I’m in the RH lane of a highway passing an offramp, and it takes a moment to figure out how to keep going straight. On the other hand, I have no expectation that it is doing anything autonomous, and this is backed up by the way it gets upset if I take my hands of the steering wheel. And unlike most other cars I hear about, it isn’t content to simply touch the wheel, it wants a hand on either “ten” or “two” specifically. For my n=1 needs, it’s fantastic: When I commute during rush hour, it handles stop-and-go traffic for me, and lets me sit back and enjoy an audiobook. It does that so well that I’ve completely lost the anxiety/aggression that formerly would have had me trying to figure out which lane is moving faster and so on. Now I just relax. Yes, I am ‘still “driving,” my hands stay on the wheel and my eyes stay on the road. But I’m more relaxed, and that is more important to me right now than whether I’m relaxed while driving or relaxed while being driven. Yes, for things like construction (and a few other cases I’ve learned to handle) I disengage it or simply override its choices with the steering wheel, accelerator, or brake. But it’s a big win over old-fashioned cruise control. I never felt like cruise control lowered the stress of highway driving the way Pilot Assist does. Mind you, none of my experience changes the fact that “Pilot Assist" is nowhere close to autonomous driving. I wouldn’t even look at it and say it’s something you can iterate on until it’s autonomous. It is what it says on the tin: Something that makes my drive easier, but doesn’t replace me as a driver.
- antiterra 5y ago
  > it wants a hand on either “ten” or “two” specifically. In the airbag era, shouldn’t this be nine and three?
  - a4isms 5y ago
    I can’t speak to that, but I can tell you that the steering wheel is shaped to make it easy to hold at ten and two.
    - chromatin 5y ago
tedmiston 5y ago
The LKAS is my Honda (Honda Sensing) has the same issue in construction zones and also anywhere there are overlapping or "ambiguous" lane markings.
barbazoo 5y ago
I wonder if the presence of autonomously driving cars will change the design of our infrastructure. The way construction area are designed nowadays expects a lot of the drivers, especially road markings. Standardizing how construction sites design their temporary road changes might make it easier for manufacturers of cars to be able to handle them. Even just proper lane markings or perhaps improved paint, developed together with manufacturers might make this issue go away.
Vik1ng 5y ago
EQS seems to do pretty well in construction areas. https://youtu.be/csdlHuUTrds?t=1549 https://youtu.be/csdlHuUTrds?t=1549

theta_d 5y ago

My 2021 Honda Accord does the same. Also turned it off.

LightG 5y ago
And ... this is why I've always said it's either all or nothing and why FSD and similar are pointless until in, who knows, 10 years time(if that) it becomes ready for prime time. Sure, youtubers can video it for fun and drink the kool aid but actual drivers will just turn it off after a hairy situation. Well, I will anyway. Maybe this is why Tesla are touting their ridiculous robot ... it's a potential pivot or alternative use of the (no doubt) impressive AI tech behind FSD which, they've now realised, will take much longer to be properly driver ready. Maybe FSD's fate is similar to the space program. The eventual secondary technology it spawns is more useful than its original purpose. Because I won't be going near these autonomous driving solutions for many years. Yeah, give me a Dalek-style home robot instead for now.

ajross 5y ago

As far as I can tell, this Drive Pilot thing is precisely the existing lane assist product. What's changed is that they've heavily constrained it so that it disengages if you go over 60 kph, or outside the hand-maintained list of blessed areas. Presumably they feel they can detect and disallow construction areas and other complexities fast enough to avoid accidents. Which is to say... this is mostly a stunt. Teslas are literally driving people around now[1], and the rest of the industry feels they need to do something. Announcing a "SAE Level 3" product, no matter how constrained, at least gets them marketing hits like this that look like an advantage. [1] No joke: mine takes my kids to and from school reliably. FSD beta isn't finished, but it's really, really good. There remain some path planning and confidence glitches that force me to disengage every dozen miles or so, but in 500 miles since I got it I've yet to see the car attempt anything genuinely unsafe. Mostly it just annoys other drivers by refusing to enter traffic.

resoluteteeth 5y ago
If Mercedes-Benz is able to define conditions where their self driving software is guaranteed to work fully autonomously without human intervention and Tesla can't, that in itself is an actual advantage over Tesla's system even if Tesla's system can do more with human monitoring in other conditions. If Tesla is also able to get approval for this (which might be hard if they push OTA updates that completely change the way the software works all the time and can't guarantee there won't be regressions) then they should by all means do so.
- ajross 5y ago
  That's sort of true in the abstract, but misses the point. Autonomy isn't about "defining conditions", it's about solving real problems. Cars that are L3 only in traffic jams on the Autobahn aren't very autonomous, nor useful. If it's stop go and the system gets above 60kph, it cuts out. If the stoppage is due to construction, it won't work. I mean, sure, it's an advantage for people who actually need to drive in exactly those conditions. But it's not a engineering advantage of the system. Fast forward a year and look at where Daimler will be and where Tesla will be. Who's going to get to a L3 system on general roads first? Who's going to be running L4 driverless vehicles first? Which is to repeat: it's just a stunt. It's a way to "sound like" they're "ahead" in this area of technology, when clearly they aren't. In fact this car is doing more or less exactly what Teslas were doing in the first version of Autopilot more than six years ago. And note that in all that time, those Teslas haven't been hitting anyone in slow traffic jams on well-maintained limited access highways either.
  - resoluteteeth 5y ago
    > That's sort of true in the abstract, but misses the point. Autonomy isn't about "defining conditions", it's about solving real problems. Cars that are L3 only in traffic jams on the Autobahn aren't very autonomous, nor useful. If it's stop go and the system gets above 60kph, it cuts out. If the stoppage is due to construction, it won't work. > I mean, sure, it's an advantage for people who actually need to drive in exactly those conditions. > But it's not a engineering advantage of the system. Fast forward a year and look at where Daimler will be and where Tesla will be. Who's going to get to a L3 system on general roads first? Who's going to be running L4 driverless vehicles first? > Which is to repeat: it's just a stunt. It's a way to "sound like" they're "ahead" in this area of technology, when clearly they aren't. In fact this car is doing more or less exactly what Teslas were doing in the first version of Autopilot more than six years ago. And note that in all that time, those Teslas haven't been hitting anyone in slow traffic jams on well-maintained limited access highways either. If it's a race to see which company can develop fully self-driving cars that work in all conditions in the next 5 years than maybe Tesla could be "ahead" of Mercedes-Benz (although it seems to be way behind other companies like Waymo). However, if we are actually decades away from fully self-driving cars that work in all conditions it is much more useful to have cars that can be trusted to operate without human supervision in limited but well defined conditions, because it is much more useful to be able to have a car that allows you to do something else other than focus on driving SOME of the time than a car that assists you more in some ways but requires supervision ALL of the time. If Tesla's approach precludes them from making self driving technology work reliably under limited conditions like this, it doesn't matter whether their self driving technology works better (but not reliably) in other conditions, because in practice it won't be as useful as more limited but reliable self driving systems. Some countries are already requiring the approach Mercedes-Benz is taking here for regulatory approval and based on the number of accidents Tesla vehicles have been involved in its possible that the US could adopt the same approach. If this happens it will not be useful at all to have a more advanced but incomplete and unreliable self-driving technology in the short term (but perhaps it will be possible to keep developing it without shipping it in cars until it reaches a sufficient level of reliability).

globular-toast 5y ago

It's more likely to kill a worker than you.

nradov 5y ago

Report it as a safety defect to the NHTSA (or your national regulator depending where you are). If enough people file similar reports then they do investigate and require manufacturers to recall the vehicles. https://www.nhtsa.gov/report-a-safety-problem https://www.nhtsa.gov/report-a-safety-problem

chollida1 5y ago

So does my Tesla:( I always take control when entering any construction site. I don't think any other than waymo has solved this yet. My telsa is good but the amount of times I have to take control from it is way too high. Snow is another area where autonomous driving sucks at the moment, and as a Canadian, man that makes my Tesla's self driving not useful 5 months of the year. Waymo seems to be way out front and everyone else seem to be fighting for second here. Though I'm very happy that we're atleast at this stage and not where we were 10 years ago. I just view it as we now have working lane keeping and adaptive cruise control for most new cars these days and that's a solid win.

ra7 5y ago
> Waymo seems to be way out front and everyone else seem to be fighting for second here. I'd say Cruise is comfortably second. They've made very good progress in SF and I believe they are doing driverless rides there for employees.
- rconti 5y ago
  That's correct, Cruise is doing driverless rides for employees in SF. I didn't know about Waymo's driverless rides, though-- can't say I follow the space that closely. Wow!
- sieabahlpark 5y ago
  Isn't that the company that stole IP from waymo?
cycrutchfield 5y ago
Just curious, what is your personal experience with Waymo that leads you to this conclusion? Also curious to know about Waymo’s performance in the snow because I haven’t heard much about that.
- nopzor 5y ago
  just took a waymo fully autonomous (no safety driver, nobody in the driver seat, no manual override) taxi in scottsdale to lunch. was surreal. supposedly over 1m miles with no faults.

ct520 5y ago

My new ram limited does the same thing

a4isms 5y ago

I use “Pilot Assist” in a 2018 Volvo V90CC, and on the one hand, it has all kinds of “issues” that require my active attention, including confusion about construction areas, or even just a general wandering when I’m in the RH lane of a highway passing an offramp, and it takes a moment to figure out how to keep going straight. On the other hand, I have no expectation that it is doing anything autonomous, and this is backed up by the way it gets upset if I take my hands of the steering wheel. And unlike most other cars I hear about, it isn’t content to simply touch the wheel, it wants a hand on either “ten” or “two” specifically. For my n=1 needs, it’s fantastic: When I commute during rush hour, it handles stop-and-go traffic for me, and lets me sit back and enjoy an audiobook. It does that so well that I’ve completely lost the anxiety/aggression that formerly would have had me trying to figure out which lane is moving faster and so on. Now I just relax. Yes, I am ‘still “driving,” my hands stay on the wheel and my eyes stay on the road. But I’m more relaxed, and that is more important to me right now than whether I’m relaxed while driving or relaxed while being driven. Yes, for things like construction (and a few other cases I’ve learned to handle) I disengage it or simply override its choices with the steering wheel, accelerator, or brake. But it’s a big win over old-fashioned cruise control. I never felt like cruise control lowered the stress of highway driving the way Pilot Assist does. Mind you, none of my experience changes the fact that “Pilot Assist" is nowhere close to autonomous driving. I wouldn’t even look at it and say it’s something you can iterate on until it’s autonomous. It is what it says on the tin: Something that makes my drive easier, but doesn’t replace me as a driver.

antiterra 5y ago
> it wants a hand on either “ten” or “two” specifically. In the airbag era, shouldn’t this be nine and three?
- a4isms 5y ago
  I can’t speak to that, but I can tell you that the steering wheel is shaped to make it easy to hold at ten and two.
  - chromatin 5y ago

tedmiston 5y ago

The LKAS is my Honda (Honda Sensing) has the same issue in construction zones and also anywhere there are overlapping or "ambiguous" lane markings.

barbazoo 5y ago

I wonder if the presence of autonomously driving cars will change the design of our infrastructure. The way construction area are designed nowadays expects a lot of the drivers, especially road markings. Standardizing how construction sites design their temporary road changes might make it easier for manufacturers of cars to be able to handle them. Even just proper lane markings or perhaps improved paint, developed together with manufacturers might make this issue go away.

Vik1ng 5y ago

EQS seems to do pretty well in construction areas. https://youtu.be/csdlHuUTrds?t=1549 https://youtu.be/csdlHuUTrds?t=1549