I don't think the vehicle performance really matters in the typical case. They're using like 20% of what the vehicle "can" do. They're probably hedging against the long tail of variance on the road somehow. Kinda like how private people can tow whatever the f they want with their pickups but in a work setting you need to keep it fairly stupid proof.

I suppose owners will be motivated to have the thing do the driving (and so seek defeat devices and such), but at least the software can have "do nothing" as a safety mode if it manages to detect that the vehicle is not configured as expected.

And maybe the software can be designed to be coupled to a vehicle dynamics model that can be updated.

The only thing an autonomous system should do with janky modified cars is drive them very slowly to the state police barracks for destruction.

A new version rolling out fast and starting to crash will likely kill the program altogether (like it did for some competitors).

10 years down the line, they won't have that risk.

Which to me is a really good, encouraging thing.

Overall I feel safer in a Waymo than a rideshare now and I only spent a few days being able to use Waymo...