The Center for Responsible Nanotechnology has an interesting article on some trends towards lower cost for weapon system development and faster weapon system development cycles. One aspect of the article is whether the "decide to kill" option should be left to armed robotic systems. I do not think that is as big an issue for a few reasons.
The decide function is already passed along from people to weapons in the form of mines and deadly traps. To a shorter timeframe in the form of gravity bombs and bullets after they are released.
The decide function would be less of an issue if the robot was using incapacitation style weapons. Tranquilizers, tasers etc...
Less lethal weapons are being actively studiedm, developed and deployed.
Robotic systems would have the option of being more selective and ethical than carpet bombing and minefields.