Think about what you just wrote: AUTOMOMOUS ACTION.
That would entail the abilty of the machine to think for itself, not only logically, but abstractly, and develop it's own sense of self, it's own tastes and dislikes, and eventually, be able to make it's own excuses for slaughtering the guy who built it.
In that case, it's no longer a machine, it's a human being with mechanical parts, and it too, will fight others of it's kind with whatever weapons it's demented little electronic mind can devise and be just as savage. The robot will merely replace the man until a new technology it develops does the same thing to it.
As a computer programmer (automation programming, no less) I can tell you that the day will never come when machines comepletely replace human beings ni any endeavor. After all, who prorgrams and builds the machines?
This will go down into that special hall of predictions:
Rue!
My autonomous robot will make you rue.
You'll rue.
Non sequitur.
The only thing that humans do currently that current computers generally do not (or do worse) with the software we write is algorithmic induction. That is it, and that won't be true for long. It is not that computers cannot do it, but it is a difficult and relatively obscure theoretical engineering problem that was poorly understood until relatively recently. And humans are not particularly good at this either, though it is the specific ability that allows us to design machines and software even with our limited competence. Go ahead and look it up; very few programmers even know what algorithmic induction is even though it is the theoretical elephant in the room of software engineering and computer science.
When computer software starts implementing pervasive algorithmic induction, the human brain will be on the fast track to obsolescence.