Self-driving cars have to answer every question they encounter successfully or it's a tragedy.

Sub detection involves a series of guesses with no significant penalty for being wrong - you just spend time looking again until you're sure. It's an infinitely easier problem.

Driving a car is in many ways harder than detecting subs. In particular, the car needs to make a lot of fine-grained decisions very fast in a feedback loop that affects what it sees. A sub-detecting algorithm only needs to output Option<vec3>. There's a lot of noise, yes, but I don't think it's crazy to think that can be solved by just having a honking huge model and adding more data, e.g. chemical traces, surface lidar, etc...

If you read the article you’d have seen that exact idea mentioned at the top. This concern is 30+ years in the future.