Flipped around, but potentially the same takeaways:
Turns out building a true FSD car is almost as hard as a humanoid robot. They've already got a Boston Dynamic's-like robot dog internally, but it sucks at getting around a parking lot.
i.e. set your expectations of the arrival of true FSD on the car to whenever you think Tesla can actually deliver this robot.
I think their trained data and current models for navigating are almost completely useless for a humanoid robot. They might have the awareness tracking for a certain "scale" of objects but a humanoid robot can trip on a half inch ledge a car can ignore and roll over. Let alone all the control systems for making the legs do leg stuff.
Boston Dynamics is already using modern AI technics and that shows on the huge advancements they had in recent years which, being revolutionary robots, are still kinda useless compared to "traditional" industrial automation costs and applicability.
Tesla is as close to being BD as it is close to being Apple and launching the next iPhone killer.
I think you are right. The gating part in both applications is, whether you can create an AI which is "good enough" to get a real understanding of the environment.
Turns out building a true FSD car is almost as hard as a humanoid robot. They've already got a Boston Dynamic's-like robot dog internally, but it sucks at getting around a parking lot.
i.e. set your expectations of the arrival of true FSD on the car to whenever you think Tesla can actually deliver this robot.