Once autonomous robots capable of exercising deadly force do begin to be used within war zones I think it will become clear that the idea of the ethical robot soldier is nonsense. In fact I think the use of such robots will make massacres and other unethical activities easier to prosecute, and also carry a greater degree of deniability for war criminals. As we know from everyday life, automaticity is frequently used as an excuse for poor service or bad behavior. In a battle situation it may be used to excuse acts which if committed by a human would be deemed reprehensible.

Incidentally, once you have a robot capable of autonomously making high level ethical decisions based upon its sensory data you've solved the grounding problem, and are a good way towards solving the AGI problem, in which case a technological singularity is pretty close at hand and you'll have a lot more to think about than merely military applications.