3 ms·

Tesla should follow Waymo's lead and release statistics on collisions while self-driving, and also the number of simulated counterfactual collisions avoided by

Tesla should follow Waymo's lead and release statistics on collisions while self-driving, and also the number of simulated counterfactual collisions avoided by the human taking over [1]. Based on the multiple incidents from the individual drivers cited in the article, it sounds like the numbers wouldn't be very good. 1: https://storage.googleapis.com/sdc-prod/v1/safety-report/Waymo-Public-Road-Safety-Performance-Data.pdf#page=5 https://storage.googleapis.com/sdc-prod/v1/safety-report/Way... Disclaimer: I work for Google but am not involved with Waymo. Opinions my own.

mlindner 6y ago
> Tesla should follow Waymo's lead and release statistics on collisions while self-driving Tesla does rely statistics on collisions while self-driving. https://www.tesla.com/VehicleSafetyReport https://www.tesla.com/VehicleSafetyReport > In the 3rd quarter [of 2020], we registered one accident for every 4.59 million miles driven in which drivers had Autopilot engaged. For those driving without Autopilot but with our active safety features, we registered one accident for every 2.42 million miles driven. For those driving without Autopilot and without our active safety features, we registered one accident for every 1.79 million miles driven.
- Strilanc 6y ago
  This is good, I had forgotten they released that data, but I was still asking for a lot more. For example, Waymo bucketed accidents by severity and went down to very low severity (like a pedestrian bumping into the side of a stopped car) whereas on that site Tesla is essentially only reporting airbag deployment counts. (To Tesla's credit, they have more time granularity than Waymo.) I do think the counterfactual "how many times did the human driver taking over prevent a crash" data is really important to have. In the article, the scary near misses they talked about fell into that category rather than into the "actual airbag deployment" category.
  - mlindner 6y ago
    > I do think the counterfactual "how many times did the human driver taking over prevent a crash" data is really important to have. In the article, the scary near misses they talked about fell into that category rather than into the "actual airbag deployment" category. It would be nice to have however I think it misrepresents. If the end result of having Autopilot on is that the driver, on average, crashes less then it's an improvement, even if the driver has to regularly intervene. What's important is the end effect, not how it happens.
    - Strilanc 6y ago
      Sure, but if you want to estimate how far they are from level 4 autonomy then the counterfactual data would be very useful.

mlindner 6y ago

> Tesla should follow Waymo's lead and release statistics on collisions while self-driving Tesla does rely statistics on collisions while self-driving. https://www.tesla.com/VehicleSafetyReport https://www.tesla.com/VehicleSafetyReport > In the 3rd quarter [of 2020], we registered one accident for every 4.59 million miles driven in which drivers had Autopilot engaged. For those driving without Autopilot but with our active safety features, we registered one accident for every 2.42 million miles driven. For those driving without Autopilot and without our active safety features, we registered one accident for every 1.79 million miles driven.

Strilanc 6y ago
This is good, I had forgotten they released that data, but I was still asking for a lot more. For example, Waymo bucketed accidents by severity and went down to very low severity (like a pedestrian bumping into the side of a stopped car) whereas on that site Tesla is essentially only reporting airbag deployment counts. (To Tesla's credit, they have more time granularity than Waymo.) I do think the counterfactual "how many times did the human driver taking over prevent a crash" data is really important to have. In the article, the scary near misses they talked about fell into that category rather than into the "actual airbag deployment" category.
- mlindner 6y ago
  > I do think the counterfactual "how many times did the human driver taking over prevent a crash" data is really important to have. In the article, the scary near misses they talked about fell into that category rather than into the "actual airbag deployment" category. It would be nice to have however I think it misrepresents. If the end result of having Autopilot on is that the driver, on average, crashes less then it's an improvement, even if the driver has to regularly intervene. What's important is the end effect, not how it happens.
  - Strilanc 6y ago
    Sure, but if you want to estimate how far they are from level 4 autonomy then the counterfactual data would be very useful.