What kind of "leapfrog" do you think is necessary to produce a "killer fully autonomous robot"?
We've actually had "autonomous killer robots," machines that kill based on the outcome of some sensor plus processing, for centuries, and fairly sophisticated ones have been feasible for decades. (For example it's trivial to build a "killer robot" that triggers off of face or voice recognition.)
The only thing that's changed recently is the kind of decisions and how effectively the robot can go looking for someone to kill.