Are Waymo's vehicles capable of autonomously driving more safely than an average human?

I personally commissioned this paper. Unfortunately, the conclusion is inconclusive.

Freely available PDF:

https://drive.google.com/file/d/1ZuJ1vA_ghf2thelqYiTfZEldVhmvhU35/view?usp=drivesdk

Official link for the pre-print (not paywalled but will bug you for an email address):

https://www.researchgate.net/publication/351450811_Are_Waymo's_vehicles_capable_of_autonomously_driving_more_safely_than_an_average_human

Interesting. In a similar vein - I recently downloaded the latest NHTSA CRSS crash data database and spent a couple of days going through it with an eye to any kind of statements that could be made about the safety of Autopilot. The data is a sampling of all the police reported crashes in the United States and includes 54k events involving 94k motor vehicles. At the end of it all I concluded that you couldn’t make statistically meaningful statements based on the information in the database. 89 events involving Tesla vehicles are present in the database, but these involve vehicles with a variety of different capabilities which are not captured in the data aside from model year information and vehicle VIN number. I looked at Tesla’s relative incidence after filtering for various criteria, but comparisons between Tesla vehicles and between Tesla and comparable competitive vehicles are difficult given the lack of ‘miles travelled’ data - which I cannot find from any good source.

The data I examined was for calendar year 2019. The 2020 data should be out soon and it might offer more samples to choose from as Tesla’s fleet size continues to grow and the mix shifts to a greater fraction of ADAS equipped vehicles. Additionally, the availability and application of ADAS is now being recorded in police reports - such data was very sparse in 2019 but should be much better captured in the 2020 data which should be out soon.

1 Like