add more inaccurate sensors and do corrections in software? We have photos of black holes out of almost a pure noise, surely it is possible to build self driving car with current gen of hardware.
The point here is that Tesla have already sold the $10k FSD sensor pack to people and told them that one day in the future a software update will provide FSD, with no additional sensors necessary. Adding more sensors is not an option (well, maybe Tesla will do that for people who bought the upgrade, but it seems unlikely they'd do that for free.)
One part of the argument you are missing is that FSD requires real-time decision making as well, so when complexity goes up with a larger number of inaccurate sensors that can become an issue. No one cares if getting a picture of a black hole from noisy data takes one year to process, the same is not true for self driving.
