3 ms·

I'm probably in the minority in the HN audience that thinks this... but we are a long way off from fully self driving cars that can be used anywhere across the

I'm probably in the minority in the HN audience that thinks this... but we are a long way off from fully self driving cars that can be used anywhere across the US. Be it Waymo, Tesla, or Cruise the problem is that these systems have way too many edge cases. For example, what happens when the sun is "blinding" you and prevents identification of the traffic light? What happens when there's construction and a construction worker gives confusing instructions to follow? This isn't unique to Tesla-- I've seen waymo cars nearly force a cyclist into a car door in broad daylight. In Tesla's case they've also been particularly aggressive with releasing self-driving tech to the end user without much stipulation on how safe it really is. I wonder how many more easily preventable deaths will have to follow before people crack down on such lax rollout of autonomous vehicle tech.

giraffe_lady 5y ago
I don't think the self-driving technology is good enough, agreed. But the legal and cultural practices they develop to fit these into already existing frameworks and expectations is also technology. I think they're likely to have enough expertise at these other skills to enable a national rollout of self-driving cars even if they aren't safe enough. And after all, we have a lot of practice ignoring road deaths and a lot of legal precedent placing that burden anywhere but on the car manufacturers. I don't think a handful of preventable deaths will create any real obstacles. Like I said we already have a lot of those. Plus as usual the people most likely to be hurt are going to be non-car users in pedestrian-hostile places, which are also harms we don't really prioritize addressing. I also think, and someone is very likely to bring it up here, the "WELL BY THAT LOGIC" argument does a lot of rhetorical favors for the self driving cars. Any argument you can make about self-driving cars being dangerous is also likely an argument about cars being dangerous. Which is true, but if you acknowledge it you get discredited as a universal car hater. Anyway there are a lot of good reasons not to allow automated cars on public roads right now. Unfortunately I don't think any of those arguments will stop them.
ModernMech 5y ago
> What happens when there's construction and a construction worker gives confusing instructions to follow? Indeed and I believe this is one of the primary failings of the nascent self-driving car industry; driving is seen as an engineering challenge, rather than an engineering and social challenge. When viewed solely as an engineering challenge, the task is daunting enough, but maybe tractable if you squint and target only a subset of driving conditions. But add in the social component of driving, which is largely being ignored by this new industry, and it's really impossible with today's technology.
- dml2135 5y ago
  Another example I saw on HN before: What happens when the tech is good enough to 100% avoid sudden pedestrian collisions, so pedestrians wise up and realize they can just jaywalk in front of any self driving car with impunity? Especially in dense areas like NYC, you can solve the safety problem, but that may require these cars to drive too passively to ever get anywhere.
  - mywittyname 5y ago
    Perhaps that will the push needed to finally separate pedestrian and vehicle traffic. Even in NYC, pedestrians are largely treated as meat sponges who are going to lose the fight with a vehicle. It's much worse in other, less walkable cities. When you put the shoe on the other foot, and give the power to the pedestrian, drivers will absolutely push to get more places to drive where pedestrians aren't allowed.
TameAntelope 5y ago
I think you get a lot of people trying to categorize the number of "years" before we get a given technology, but it's probably more accurate to measure the number of "unsolved problems", and my totally unqualified opinion is that there are still a lot of unsolved problems left before we get level 5 AV.
dogma1138 5y ago
I don’t think you are it’s clearly the case. I also think MobilEyes approach is far better than Tesla’s and they might actually win the race even if they are hardly competing in it openly. I do like the idea behind their safety model and the approach that the difference between L3+/4 and L5 is about reducing the MTBF to essentially 0 rather than just adding more compute and features. They also are probably the only company that A) has had multiple generations of their hardware and software in a lot of different types of vehicles on the road for over a decade and B) don’t undersell the problem and oversell their solutions at all. As far as I’m concerned they seem to be the only responsible player in the market right now.
throwawaylinux 5y ago
In my opinion, "AI" is still far closer to being able to recognize a picture of a cat than it is able to cope with all the complexities of driving that a person might encounter and deal with quite easily. What might be a lot closer is a restricted kind of driving that might cover 90% of people's usual cases. Restricted set of roads (well mapped and maintained), maybe specific instructions for problem sections, markings and/or signals that the car understands, maybe local updates for changed road conditions, that kind of thing.
- user_named 5y ago
  We still have to train Google to recognize images of busses with captchas.

giraffe_lady 5y ago

I don't think the self-driving technology is good enough, agreed. But the legal and cultural practices they develop to fit these into already existing frameworks and expectations is also technology. I think they're likely to have enough expertise at these other skills to enable a national rollout of self-driving cars even if they aren't safe enough. And after all, we have a lot of practice ignoring road deaths and a lot of legal precedent placing that burden anywhere but on the car manufacturers. I don't think a handful of preventable deaths will create any real obstacles. Like I said we already have a lot of those. Plus as usual the people most likely to be hurt are going to be non-car users in pedestrian-hostile places, which are also harms we don't really prioritize addressing. I also think, and someone is very likely to bring it up here, the "WELL BY THAT LOGIC" argument does a lot of rhetorical favors for the self driving cars. Any argument you can make about self-driving cars being dangerous is also likely an argument about cars being dangerous. Which is true, but if you acknowledge it you get discredited as a universal car hater. Anyway there are a lot of good reasons not to allow automated cars on public roads right now. Unfortunately I don't think any of those arguments will stop them.

ModernMech 5y ago

> What happens when there's construction and a construction worker gives confusing instructions to follow? Indeed and I believe this is one of the primary failings of the nascent self-driving car industry; driving is seen as an engineering challenge, rather than an engineering and social challenge. When viewed solely as an engineering challenge, the task is daunting enough, but maybe tractable if you squint and target only a subset of driving conditions. But add in the social component of driving, which is largely being ignored by this new industry, and it's really impossible with today's technology.

dml2135 5y ago
Another example I saw on HN before: What happens when the tech is good enough to 100% avoid sudden pedestrian collisions, so pedestrians wise up and realize they can just jaywalk in front of any self driving car with impunity? Especially in dense areas like NYC, you can solve the safety problem, but that may require these cars to drive too passively to ever get anywhere.
- mywittyname 5y ago
  Perhaps that will the push needed to finally separate pedestrian and vehicle traffic. Even in NYC, pedestrians are largely treated as meat sponges who are going to lose the fight with a vehicle. It's much worse in other, less walkable cities. When you put the shoe on the other foot, and give the power to the pedestrian, drivers will absolutely push to get more places to drive where pedestrians aren't allowed.

TameAntelope 5y ago

I think you get a lot of people trying to categorize the number of "years" before we get a given technology, but it's probably more accurate to measure the number of "unsolved problems", and my totally unqualified opinion is that there are still a lot of unsolved problems left before we get level 5 AV.

dogma1138 5y ago

I don’t think you are it’s clearly the case. I also think MobilEyes approach is far better than Tesla’s and they might actually win the race even if they are hardly competing in it openly. I do like the idea behind their safety model and the approach that the difference between L3+/4 and L5 is about reducing the MTBF to essentially 0 rather than just adding more compute and features. They also are probably the only company that A) has had multiple generations of their hardware and software in a lot of different types of vehicles on the road for over a decade and B) don’t undersell the problem and oversell their solutions at all. As far as I’m concerned they seem to be the only responsible player in the market right now.

throwawaylinux 5y ago

In my opinion, "AI" is still far closer to being able to recognize a picture of a cat than it is able to cope with all the complexities of driving that a person might encounter and deal with quite easily. What might be a lot closer is a restricted kind of driving that might cover 90% of people's usual cases. Restricted set of roads (well mapped and maintained), maybe specific instructions for problem sections, markings and/or signals that the car understands, maybe local updates for changed road conditions, that kind of thing.

user_named 5y ago
We still have to train Google to recognize images of busses with captchas.