inertia (IMU), vision (CCD), skin pressure (not sure?), absolute offset/absolute rotation (optical encoders)
so now the question is: how do we convert this bag of signals into mimicking a human. this question has been asked for text already, and the answer is LLMs which can, at the very least, mimick humans pretty well.
if a humanoid can be made to mimick a cook, or a cleaner, or manual labor, be able to navigate human geometry (stairs, ramps, etc.) that is already huge.
i agree that it would be the most advanced consumer grade product - the only thing that might beat it, is a domestic nuclear reactor.