7 ms·

That is actually an interesting question. We track how many accidents the automated cars cause, but I'm not sure how many accidents they prevent are tracked.

jimnotgym 3y ago
They should let an accident happen occasionally, and then show how it had the data to avoid it if humans had just trusted them.
olex 3y ago
That metric is tracked indirectly, by the lower number of accidents that AVs are involved in per distance travelled compared to human-driven vehicles.
- reneherse 3y ago
  It would be interesting if some kind of active/inductive measurement could be made. As a human driver, I'm keenly aware of the "close calls" I've had where I've narrowly avoided a probable collision through good decisions and/or reaction times. I'm probably less aware of the times when I've just gotten lucky. No doubt self-driving companies have internally available data on this stuff. Can't wait til superior performance to human drivers is a major selling point!
- delfinom 3y ago
  I still want that metric restricted to geographical area. The average across the entire country seems many times outright malicious to use.
- CSMastermind 3y ago
  > by the lower number of accidents that AVs are involved in per distance travelled compared to human-driven vehicles Just a note that this number is hard to calculate accurately with an acceptable degree of certainty. Anyone claiming that AVs are involved in fewer accidents per distance traveled than human drivers is either extrapolating from incomplete data or baking some unreliable assumptions into their statement. Welch Labs has a good introductory video to this topic: https://youtu.be/yaYER2M8dcs?si=XEB4aWlYf6gnnTqM https://youtu.be/yaYER2M8dcs?si=XEB4aWlYf6gnnTqM
  - Steltek 3y ago
    Can someone summarize the video? That was my first thought as well: crash data for humans is clearly underreported. For example, police don't always write reports or human drivers agree to keep it off the books.
    - serial_dev 3y ago
      And I'm not even sure how reliable is the "miles driven" metric. I mean I'm sure you can estimate it somehow, but what's the margin of error there?
      - viraptor 3y ago
        Odometers are pretty well regulated and insurance companies will often have a good record of the readings over long periods. I'm not sure how the org doing the data collection does it precisely, but pretty accurate data is out there.
    - CSMastermind 3y ago
      The probability of a human driver causing a fatality on any given mile driven is 0.00000109% (1.09 fatalities occur per 100 million miles driven). Applying some basic statistics to show to a 95% confidence level that a self driving system causes fewer fatalities than a human you would have to drive 275 million autonomous miles flawlessly. This would require a fleet of 100 vehicles to drive continuously for 12.56 years. And in practice self-driving vehicles don't drive perfectly. Best estimates on the actual number of miles driven to validate their safety is around 5 billion autonomously driven miles and that's assuming that they actually are safer than a human driver. Then you get into the comparison itself. In practice AVs don't drive on all the same roads, at the same times, as human drivers. A disproportionate number of accidents happen at night, in adverse weather, and on roads that AVs don't typically drive on. Then you have to ask if comparing AVs to all drivers and vehicles is valid comparison. We know for instance that vehicles with automated breaking and lane assist are involved in fewer accidents. Then of course if minimizing accidents is really what you care about there's something easy we could do right now: just mandate all vehicles must have an breathalyzer ignition. We do this for some people who have been convicted of DUI but doing it for everyone would eliminate a third of fatalities.
- Mathnerd314 3y ago
  Not sure that's fair - the number of accidents in human-driven vehicles varies significantly by the human driving it. Do you compare with the teen that just obtained their license and is more interested in their phone than in driving properly, or the 65-year old doctor who's been driving for work and other purposes every day for the past 40 years? According to https://injuryfacts.nsc.org/motor-vehicle/overview/age-of-driver/ https://injuryfacts.nsc.org/motor-vehicle/overview/age-of-dr... it's at least a 6x difference.
  - saagarjha 3y ago
    Teens who’ve just gotten their license are often quite good drivers. It’s after that when they drive poorly. (This also applies to adults, to a lesser extent.)
  - hcurtiss 3y ago
    I think you compare it to the average as that's who you are encountering when on the road. You have no way of selecting the other drivers to bias toward the doctor.
  - a_wild_dandan 3y ago
    You compare what matters — mean accidents per mile driven. And like any actuary, you can compare distributions with finer granularity data (location, tickets, ages, occupations, collusion severity, etc). None of this is new or intractable. We can have objective standards with confident judgements of what’s safe for our roads. As an aside, public acceptance of driverless cars is another story. A story filled with warped perspectives due to media outlets stoking the audience emotions which keep outlets in business — outrage, fear, and drama. For every story covering an extreme accident involving human drivers, there might be 100 stories covering a moderate driverless accident. No matter how superhuman autonomous cars perform, they’ll have a massive uphill battle for a positive image.
- stouset 3y ago
  Although things are changing, the overwhelming majority of AV miles are generally-safer miles on protected highways. And yet their statistics are typically compared against vehicle miles on all road types. Further, most AVs explicitly disengage when they encounter situations beyond their understanding. So again, we’re selecting AV statistics based on generally favorable conditions, but we don’t track human-operated miles the same way. It’s not really a fair comparison.
  - l33t7332273 3y ago
    > most AVs explicitly disengage when they encounter situations beyond their understanding What do you mean by disengage? Cruise’s AVs don’t have a driver to take over.
    - stouset 3y ago
      I’m lumping things like Tesla FSD. But also, there are explicitly times and areas where Cruises don’t operate due to being insufficiently able to operate. And times where they do just pull over and wait for human intervention when they don’t know what else to do. Both of which are safe and reasonable, but which also selects themselves into a safer set of miles actually traveled.
      - Zanni 3y ago
        Human drivers "disengage" also, though we don't think of it that way. I have aging friends who refuse to drive at night. I'll stay home in bad weather. When I was younger, I'd sometimes ask a passenger to back out of tricky parking space for me.
        stouset 3y ago
        Yes, but we’re still comparing a highly selected set of AV miles vs the massive variety of human-driven miles.
  - xnx 3y ago
    It's not too useful to lump Tesla, Cruise, and Waymo together here. Tesla is years behind Cruise and cruise is years behind Waymo in terms of driving capability. Waymo doesn't even drive on highways, so we don't know how safe it would be (probably very safe).
- Guy_w_Keyboard 3y ago
  Well, obviously the number of autonomous vehicles involved in accidents is going to be lower, but that's because barely any of them exist compared to the vast majority of people driving their cars. If you had statistics on proportions though, that might be a different story.
  - viraptor 3y ago
    You missed "per distance travelled" - that partially normalises the results. You still have to adjust for situation and road type (which Tesla conveniently doesn't do in their advertised numbers) for a better result.
dron57 3y ago
The companies do track that data internally. It would be nice if the DMV mandated it's release to the public. But of course this type of data is a counterfactual, so much more subjective.
thesz 3y ago
The same applies to regular drivers. We do not track accidents they prevent.
- throw310822 3y ago
  Easy, humans prevent nearly 100% of the accidents. A car without a driver invariably crashes in a few seconds. The question is how many accidents (if any) self driving prevents with respect to average human drivers.
  - bumby 3y ago
    This is such an odd take I don’t know if it’s trolling. Both can be measured against similar metrics (an AV model against the avg AV/avg human or vice versa.
  - thesz 3y ago
    Self-driving cars are used in a very controlled environment. They will not function in high grass, I guess from my experience with different parktronics. The best "off road" demonstration of self-driving and/or AI assistance I've found is this: https://media.jaguarlandrover.com/news/2016/07/jaguar-land-rover-demonstrates-all-terrain-self-driving-technology https://media.jaguarlandrover.com/news/2016/07/jaguar-land-r... Note they avoid going into grass. Human can deduce trails from grass profile, AI? I don't think so. Will you count an inability to reach a lakeside as an accident? I guess, no.
    - RecycledEle 3y ago
      We have not trained AIs to deduce trails from grass. AIs need training before they can do things.* *They just might be learning to do things without being trained based on the emergency behavior I see in LLMs.
      - thesz 3y ago
        My point was not about discerning track from grass, but about driving in acttual grass. Current self-driving tech uses sensors that are useless in high grass. As for "emergency behavior" (emergent behavior, I guess) - we do not know how LLMs are trained. Thus, what you consider "emergency behavior," could very well be a part of training.
fiddlerwoaroof 3y ago
I’ve had several low-light situations where my Tesla identified a pedestrian and applied the brakes before I did (typically dark clothing on a dark street).
- ajross 3y ago
  Ditto, FWIW. The car may not always behave like a human does in those circumstances[1], but at this point it's objectively much better at the attention and recognition side of the task. [1] It tends to be more conservative than human drivers, and in weird ways. If a pedestrian seems at all to be moving in the direction of traffic, even if they're on a sidewalk and just meandering a bit, the car will try to evade (not violently, but it will brake a bit and pull to the outside of the lane).
  - fiddlerwoaroof 3y ago
    Yeah, if anything, my Tesla’s issues seem to stem from being overly concerned about accidents than being recklessly dangerous.
- mikeryan 3y ago
  This is low hanging fruit I had this feature in my 2015 Volvo.
  - fiddlerwoaroof 3y ago
    Sure, it’s just more evidence that these automated systems improve overall safety vs. an unassisted driver. (Although I worry a bit about automation-induced inattention negating these benefits)
    - lemming 3y ago
      There are also issues like phantom braking which Teslas are prone to (or were, I'm not sure if that's better these days). That's part of a whole class of problems which the AVs suffer from which humans don't. I think the main problem is that those problems are really unpredictable to human drivers, whereas good defensive drivers will take into account what another lousy human driver might do based on lots of experience.
jsemrau 3y ago
Maybe of interest. There has been research by SwissRe on this topic. https://jdsemrau.substack.com/p/paper-review-comparative-safety-performance https://jdsemrau.substack.com/p/paper-review-comparative-saf...
- twoWhlsGud 3y ago
  That's behind a paywall. Do you have a link to the SwissRe paper by chance? (Which might also behind a paywall, of course ; )
  - microtherion 3y ago
    Seems to be this one: https://arxiv.org/ftp/arxiv/papers/2309/2309.01206.pdf https://arxiv.org/ftp/arxiv/papers/2309/2309.01206.pdf
  - xnx 3y ago
    Waymo also talks about its work with SwissRe here: https://www.youtube.com/watch?v=9-Qu6HNZu8g https://www.youtube.com/watch?v=9-Qu6HNZu8g
xnx 3y ago
Waymo thinks about this a lot and has posted a good video about it here: https://www.youtube.com/watch?v=9-Qu6HNZu8g https://www.youtube.com/watch?v=9-Qu6HNZu8g

jimnotgym 3y ago

They should let an accident happen occasionally, and then show how it had the data to avoid it if humans had just trusted them.

olex 3y ago

That metric is tracked indirectly, by the lower number of accidents that AVs are involved in per distance travelled compared to human-driven vehicles.

reneherse 3y ago
It would be interesting if some kind of active/inductive measurement could be made. As a human driver, I'm keenly aware of the "close calls" I've had where I've narrowly avoided a probable collision through good decisions and/or reaction times. I'm probably less aware of the times when I've just gotten lucky. No doubt self-driving companies have internally available data on this stuff. Can't wait til superior performance to human drivers is a major selling point!
delfinom 3y ago
I still want that metric restricted to geographical area. The average across the entire country seems many times outright malicious to use.
CSMastermind 3y ago
> by the lower number of accidents that AVs are involved in per distance travelled compared to human-driven vehicles Just a note that this number is hard to calculate accurately with an acceptable degree of certainty. Anyone claiming that AVs are involved in fewer accidents per distance traveled than human drivers is either extrapolating from incomplete data or baking some unreliable assumptions into their statement. Welch Labs has a good introductory video to this topic: https://youtu.be/yaYER2M8dcs?si=XEB4aWlYf6gnnTqM https://youtu.be/yaYER2M8dcs?si=XEB4aWlYf6gnnTqM
- Steltek 3y ago
  Can someone summarize the video? That was my first thought as well: crash data for humans is clearly underreported. For example, police don't always write reports or human drivers agree to keep it off the books.
  - serial_dev 3y ago
    And I'm not even sure how reliable is the "miles driven" metric. I mean I'm sure you can estimate it somehow, but what's the margin of error there?
    - viraptor 3y ago
      Odometers are pretty well regulated and insurance companies will often have a good record of the readings over long periods. I'm not sure how the org doing the data collection does it precisely, but pretty accurate data is out there.
  - CSMastermind 3y ago
    The probability of a human driver causing a fatality on any given mile driven is 0.00000109% (1.09 fatalities occur per 100 million miles driven). Applying some basic statistics to show to a 95% confidence level that a self driving system causes fewer fatalities than a human you would have to drive 275 million autonomous miles flawlessly. This would require a fleet of 100 vehicles to drive continuously for 12.56 years. And in practice self-driving vehicles don't drive perfectly. Best estimates on the actual number of miles driven to validate their safety is around 5 billion autonomously driven miles and that's assuming that they actually are safer than a human driver. Then you get into the comparison itself. In practice AVs don't drive on all the same roads, at the same times, as human drivers. A disproportionate number of accidents happen at night, in adverse weather, and on roads that AVs don't typically drive on. Then you have to ask if comparing AVs to all drivers and vehicles is valid comparison. We know for instance that vehicles with automated breaking and lane assist are involved in fewer accidents. Then of course if minimizing accidents is really what you care about there's something easy we could do right now: just mandate all vehicles must have an breathalyzer ignition. We do this for some people who have been convicted of DUI but doing it for everyone would eliminate a third of fatalities.
Mathnerd314 3y ago
Not sure that's fair - the number of accidents in human-driven vehicles varies significantly by the human driving it. Do you compare with the teen that just obtained their license and is more interested in their phone than in driving properly, or the 65-year old doctor who's been driving for work and other purposes every day for the past 40 years? According to https://injuryfacts.nsc.org/motor-vehicle/overview/age-of-driver/ https://injuryfacts.nsc.org/motor-vehicle/overview/age-of-dr... it's at least a 6x difference.
- saagarjha 3y ago
  Teens who’ve just gotten their license are often quite good drivers. It’s after that when they drive poorly. (This also applies to adults, to a lesser extent.)
- hcurtiss 3y ago
  I think you compare it to the average as that's who you are encountering when on the road. You have no way of selecting the other drivers to bias toward the doctor.
- a_wild_dandan 3y ago
  You compare what matters — mean accidents per mile driven. And like any actuary, you can compare distributions with finer granularity data (location, tickets, ages, occupations, collusion severity, etc). None of this is new or intractable. We can have objective standards with confident judgements of what’s safe for our roads. As an aside, public acceptance of driverless cars is another story. A story filled with warped perspectives due to media outlets stoking the audience emotions which keep outlets in business — outrage, fear, and drama. For every story covering an extreme accident involving human drivers, there might be 100 stories covering a moderate driverless accident. No matter how superhuman autonomous cars perform, they’ll have a massive uphill battle for a positive image.
stouset 3y ago
Although things are changing, the overwhelming majority of AV miles are generally-safer miles on protected highways. And yet their statistics are typically compared against vehicle miles on all road types. Further, most AVs explicitly disengage when they encounter situations beyond their understanding. So again, we’re selecting AV statistics based on generally favorable conditions, but we don’t track human-operated miles the same way. It’s not really a fair comparison.
- l33t7332273 3y ago
  > most AVs explicitly disengage when they encounter situations beyond their understanding What do you mean by disengage? Cruise’s AVs don’t have a driver to take over.
  - stouset 3y ago
    I’m lumping things like Tesla FSD. But also, there are explicitly times and areas where Cruises don’t operate due to being insufficiently able to operate. And times where they do just pull over and wait for human intervention when they don’t know what else to do. Both of which are safe and reasonable, but which also selects themselves into a safer set of miles actually traveled.
    - Zanni 3y ago
      Human drivers "disengage" also, though we don't think of it that way. I have aging friends who refuse to drive at night. I'll stay home in bad weather. When I was younger, I'd sometimes ask a passenger to back out of tricky parking space for me.
      - stouset 3y ago
        Yes, but we’re still comparing a highly selected set of AV miles vs the massive variety of human-driven miles.
- xnx 3y ago
  It's not too useful to lump Tesla, Cruise, and Waymo together here. Tesla is years behind Cruise and cruise is years behind Waymo in terms of driving capability. Waymo doesn't even drive on highways, so we don't know how safe it would be (probably very safe).
Guy_w_Keyboard 3y ago
Well, obviously the number of autonomous vehicles involved in accidents is going to be lower, but that's because barely any of them exist compared to the vast majority of people driving their cars. If you had statistics on proportions though, that might be a different story.
- viraptor 3y ago
  You missed "per distance travelled" - that partially normalises the results. You still have to adjust for situation and road type (which Tesla conveniently doesn't do in their advertised numbers) for a better result.

dron57 3y ago

The companies do track that data internally. It would be nice if the DMV mandated it's release to the public. But of course this type of data is a counterfactual, so much more subjective.

thesz 3y ago

The same applies to regular drivers. We do not track accidents they prevent.

throw310822 3y ago
Easy, humans prevent nearly 100% of the accidents. A car without a driver invariably crashes in a few seconds. The question is how many accidents (if any) self driving prevents with respect to average human drivers.
- bumby 3y ago
  This is such an odd take I don’t know if it’s trolling. Both can be measured against similar metrics (an AV model against the avg AV/avg human or vice versa.
- thesz 3y ago
  Self-driving cars are used in a very controlled environment. They will not function in high grass, I guess from my experience with different parktronics. The best "off road" demonstration of self-driving and/or AI assistance I've found is this: https://media.jaguarlandrover.com/news/2016/07/jaguar-land-rover-demonstrates-all-terrain-self-driving-technology https://media.jaguarlandrover.com/news/2016/07/jaguar-land-r... Note they avoid going into grass. Human can deduce trails from grass profile, AI? I don't think so. Will you count an inability to reach a lakeside as an accident? I guess, no.
  - RecycledEle 3y ago
    We have not trained AIs to deduce trails from grass. AIs need training before they can do things.* *They just might be learning to do things without being trained based on the emergency behavior I see in LLMs.
    - thesz 3y ago
      My point was not about discerning track from grass, but about driving in acttual grass. Current self-driving tech uses sensors that are useless in high grass. As for "emergency behavior" (emergent behavior, I guess) - we do not know how LLMs are trained. Thus, what you consider "emergency behavior," could very well be a part of training.

fiddlerwoaroof 3y ago

I’ve had several low-light situations where my Tesla identified a pedestrian and applied the brakes before I did (typically dark clothing on a dark street).

ajross 3y ago
Ditto, FWIW. The car may not always behave like a human does in those circumstances[1], but at this point it's objectively much better at the attention and recognition side of the task. [1] It tends to be more conservative than human drivers, and in weird ways. If a pedestrian seems at all to be moving in the direction of traffic, even if they're on a sidewalk and just meandering a bit, the car will try to evade (not violently, but it will brake a bit and pull to the outside of the lane).
- fiddlerwoaroof 3y ago
  Yeah, if anything, my Tesla’s issues seem to stem from being overly concerned about accidents than being recklessly dangerous.
mikeryan 3y ago
This is low hanging fruit I had this feature in my 2015 Volvo.
- fiddlerwoaroof 3y ago
  Sure, it’s just more evidence that these automated systems improve overall safety vs. an unassisted driver. (Although I worry a bit about automation-induced inattention negating these benefits)
  - lemming 3y ago
    There are also issues like phantom braking which Teslas are prone to (or were, I'm not sure if that's better these days). That's part of a whole class of problems which the AVs suffer from which humans don't. I think the main problem is that those problems are really unpredictable to human drivers, whereas good defensive drivers will take into account what another lousy human driver might do based on lots of experience.

jsemrau 3y ago

Maybe of interest. There has been research by SwissRe on this topic. https://jdsemrau.substack.com/p/paper-review-comparative-safety-performance https://jdsemrau.substack.com/p/paper-review-comparative-saf...

twoWhlsGud 3y ago
That's behind a paywall. Do you have a link to the SwissRe paper by chance? (Which might also behind a paywall, of course ; )
- microtherion 3y ago
  Seems to be this one: https://arxiv.org/ftp/arxiv/papers/2309/2309.01206.pdf https://arxiv.org/ftp/arxiv/papers/2309/2309.01206.pdf
- xnx 3y ago
  Waymo also talks about its work with SwissRe here: https://www.youtube.com/watch?v=9-Qu6HNZu8g https://www.youtube.com/watch?v=9-Qu6HNZu8g

xnx 3y ago

Waymo thinks about this a lot and has posted a good video about it here: https://www.youtube.com/watch?v=9-Qu6HNZu8g https://www.youtube.com/watch?v=9-Qu6HNZu8g