3 ms·

> Aren't they operating under a lot of constraints? The comparison should be with human drivers operating under similar conditions and on similar roads, similar

by ra7 3y ago

> Aren't they operating under a lot of constraints? The comparison should be with human drivers operating under similar conditions and on similar roads, similar cars(eg. new ones), weather conditions etc. The insurance data for the comparison is from the zip codes Waymo currently operates in, so all of that is accounted for. Here is the full study: https://arxiv.org/pdf/2309.01206.pdf https://arxiv.org/pdf/2309.01206.pdf

notyourwork 3y ago
And for cars that are geofenced. Not accurate comparison.
- ehsankia 3y ago
  If you're comparing to other accidents in the same (geofenced) zipcode, then why is it not an accurate comparison?
  - lolinder 3y ago
    The linked study even calls this out: the zip codes they controlled for are the residence zip code, which means accidents that happen on the freeway (where Waymo typically doesn't drive) or even in other zip codes are included in the human baseline.
    - notyourwork 3y ago
      Precisely.
    - mahkeiro 3y ago
      In Phoenix, the waymo drive on the freeway around the city, and given the number of crashes I see in that city I have few doubt that Waymo cars are safer.
lolinder 3y ago
"All of that" is not encompassed in the human driver's home zip code. OP is talking about more than just the zip code, they're also talking about restrictions on the times of day Waymo operates, the types of vehicles they use, whether Waymo shuts down in adverse weather, etc. I'm not sure what all the current and past restrictions are, but those have to be controlled for. The study you linked only discusses one uncontrolled variable (which is a bad sign given how many others there are that they don't talk about) but it's one I hadn't thought of: > A limitation of the selected human baseline is that the location of crashes that generate claims is not known, which limits the ability to filter claims based on Waymo's Operational Design Domain (ODD). As a result, whereas the Waymo ODD largely does not include freeway driving, the human database includes miles driven and claims which occur on freeways. Due to variations in collision frequency per million miles between freeways and non-freeways, this may have led to a baseline which may be more conservative than a roadway- matched baseline.
- ra7 3y ago
  Fair enough. Maybe not “all of it”, but most of it is covered. Waymo has always operated 24/7 since the beginning and had 99.4% fleet uptime [1] during the past winter season (SF had record rains). They used to shut down during inclement weather in Phoenix years ago, but their SF rain operations have scaled up with the mileage. Waymo vehicles use its own set of sensors and don’t rely on the vehicle’s safety features, so I’m not sure how types of vehicles is relevant for the comparison. This is pretty good as far as “apples-to-apples” comparisons go. [1] https://waymo.com/blog/2023/08/the-waymo-drivers-rapid-learning-curve.html https://waymo.com/blog/2023/08/the-waymo-drivers-rapid-learn...
  - lolinder 3y ago
    I don't live in SF, but I was under the impression that Waymo's 24/7 operation is a relatively new thing [0]. EDIT: On further research, it looks like for Waymo this change only applied to when they could charge fares, so OP is correct that they were authorized to operate 24/7 [1]. [0] Article from August 10 of this year: https://www.theverge.com/2023/8/10/23827790/waymo-cruise-cpuc-vote-robotaxi-san-francisco https://www.theverge.com/2023/8/10/23827790/waymo-cruise-cpu... [1] https://www.cpuc.ca.gov/news-and-updates/all-news/cpuc-approves-permits-for-cruise-and-waymo-to-charge-fares-for-passenger-service-in-sf-2023 https://www.cpuc.ca.gov/news-and-updates/all-news/cpuc-appro...

by ra7 3y ago

notyourwork 3y ago

And for cars that are geofenced. Not accurate comparison.

ehsankia 3y ago
If you're comparing to other accidents in the same (geofenced) zipcode, then why is it not an accurate comparison?
- lolinder 3y ago
  The linked study even calls this out: the zip codes they controlled for are the residence zip code, which means accidents that happen on the freeway (where Waymo typically doesn't drive) or even in other zip codes are included in the human baseline.
  - notyourwork 3y ago
    Precisely.
  - mahkeiro 3y ago
    In Phoenix, the waymo drive on the freeway around the city, and given the number of crashes I see in that city I have few doubt that Waymo cars are safer.

lolinder 3y ago

"All of that" is not encompassed in the human driver's home zip code. OP is talking about more than just the zip code, they're also talking about restrictions on the times of day Waymo operates, the types of vehicles they use, whether Waymo shuts down in adverse weather, etc. I'm not sure what all the current and past restrictions are, but those have to be controlled for. The study you linked only discusses one uncontrolled variable (which is a bad sign given how many others there are that they don't talk about) but it's one I hadn't thought of: > A limitation of the selected human baseline is that the location of crashes that generate claims is not known, which limits the ability to filter claims based on Waymo's Operational Design Domain (ODD). As a result, whereas the Waymo ODD largely does not include freeway driving, the human database includes miles driven and claims which occur on freeways. Due to variations in collision frequency per million miles between freeways and non-freeways, this may have led to a baseline which may be more conservative than a roadway- matched baseline.

ra7 3y ago
Fair enough. Maybe not “all of it”, but most of it is covered. Waymo has always operated 24/7 since the beginning and had 99.4% fleet uptime [1] during the past winter season (SF had record rains). They used to shut down during inclement weather in Phoenix years ago, but their SF rain operations have scaled up with the mileage. Waymo vehicles use its own set of sensors and don’t rely on the vehicle’s safety features, so I’m not sure how types of vehicles is relevant for the comparison. This is pretty good as far as “apples-to-apples” comparisons go. [1] https://waymo.com/blog/2023/08/the-waymo-drivers-rapid-learning-curve.html https://waymo.com/blog/2023/08/the-waymo-drivers-rapid-learn...
- lolinder 3y ago
  I don't live in SF, but I was under the impression that Waymo's 24/7 operation is a relatively new thing [0]. EDIT: On further research, it looks like for Waymo this change only applied to when they could charge fares, so OP is correct that they were authorized to operate 24/7 [1]. [0] Article from August 10 of this year: https://www.theverge.com/2023/8/10/23827790/waymo-cruise-cpuc-vote-robotaxi-san-francisco https://www.theverge.com/2023/8/10/23827790/waymo-cruise-cpu... [1] https://www.cpuc.ca.gov/news-and-updates/all-news/cpuc-approves-permits-for-cruise-and-waymo-to-charge-fares-for-passenger-service-in-sf-2023 https://www.cpuc.ca.gov/news-and-updates/all-news/cpuc-appro...