Inventing an entirely new set of standard "interfaces" (used broadly) that are robot specialized, then installing them literally everywhere in place of the existing human specialized interfaces, has both developmental and deployment problems.
It's a huge upfront cost, for one. "What do you mean I need to rebuild my kitchen, laundry room, lawn shed, bedroom, bathroom, and utility rooms before I can use your home assistant robot? I'll just go buy the Honda one that has hands. Also, I kind of like cooking and would like the option of doing it myself sometimes."
And, well, at the end of the day why is the non-humanoid one better? Even granting your premise that it is more "efficient" in the context of a house designed around it (as opposed to designed around the person living in it and paying for it), as soon as you want to do something the designers failed to anticipate you are boned. If the robot is mechanically humanoid, new "tasks" can be added with just a software update (or even by having the person demonstrate what they want). If it's an eldritch monstrosity that can only operate in spaces designed around it, you are stuck unless you call a mechanic. And what happens if the robot breaks? If the house is still human adapted, you can cope. But if I have to get the robot repaired before I can operate the microwave....
Moreover, there is an issue with getting people to "trust" the robots, and having the robots be things that are more or less human looking is a good way to build trust.
These are not hypotheticals I'm spitballing, by the way. These are things robotics researchers are actively studying.