This article is bad journalism. The issue with using an NHTSA statistic to say we "avoid" crashes at a particular rate is that it only covers what it measures,

This article is bad journalism. The issue with using an NHTSA statistic to say we "avoid" crashes at a particular rate is that it only covers what it measures, which means an event was significant enough to enter one of the various agencies that are part of the survey, and that generally means "it was reported to police". It biases our assessment of ourselves positively by omission. [0] The companies do present data if you actually lift a finger to search it out. Waymo's figure in their "first million rider-only miles" report [1] has 20 collision events, or a 99.998% rate per mile. That encompasses every bump of any significance, a considerably more rigorous standard, and it is more likely to produce events-per-mile because of its urban service areas. Although they didn't attempt to turn these events into an apples-to-apples "vs NHTSA urban crash rate" figure, this is a small enough quantity that they present every one in a table for your own assessment. In the quantities that are out on the streets, the additional 9's being asked for will be noise, and a more qualitative assessment is called for anyway. [0] https://cdan.nhtsa.gov/tsftables/tsfar.htm# https://cdan.nhtsa.gov/tsftables/tsfar.htm# [1] https://waymo.com/safety/ https://waymo.com/safety/