That is absolutely correct. Figuring out when the watch is in an upright position is trivial. Figuring out whether the user intended to actually get it to turn on is extremely hard, and it involves looking at the history of motion that preceded it, and making judgments based on a whole lot of user data.

The other problem is making the tradeoff between detecting this fast enough (to turn on all the systems that follow so that the screen can light up fast enough) and waiting long enough (so you can be extra sure that the user intended to look at the screen)

The garden path is easy. Making it work mostly unobtrusively is very hard.