> Why wouldn't those same techniques be useful for more general object tracking?

These demos use machine learning to train against a known environment.

Basically, pattern matching changes in the signals against a very controlled set of training data.

You can use WiFi signals to detect that something is changing in the environment, but without the machine learning with controlled input data you don't know what it actually means. This is how WiFi presence detection works, but it won't tell you if it's a person moving through the house or your cat walked in front of the router.