(Ex-Waymo) Waymo tries really hard to compare apples to apple miles, considering a number of dimensions (weather, speed limits of the roads the miles are on, the specific cities, and more). This is one of the things that they feel distinguishes them in the space.

The data on https://waymo.com/safety/ reflects these apples to apples comparison.

They’ve also, when possible, done comparisons of the Waymo driver to an ideally attentive human driver. https://waymo.com/blog/2022/09/benchmarking-av-safety/

The biggest challenge is in sourcing the data. A lot of these datasets don't exist or have to be developed.

(The article itself, mixes and matches stats which is a common thing in general media).