Dallas Police used a bomb defusing robot to blow a guy up... so it's not too hard to imagine.
https://www.theverge.com/2016/7/8/12128230/dallas-police-bom...
I suppose one small difference would be whether the bomb disposal robot in that situation was destroyed by the explosion it caused, since it presumably wouldn't be destroyed by shooting a gun (unless the suspect returned fire or had a dead man's switch on his own bomb). I highlight this difference not because of the ethics of sending robots to their "deaths", but the minor practical consideration of whether a given robot is considered reusable, that is, able to amass a potentially unlimited number of kills.
To place a bomb is a linear bang-bang controller (pun intended) And the decision is already made wrt collateral damage
But to target a moving object from a moving object holding a gun on a 3-dof mount is an order of magnitude harder to control and the question of collateral damage is an unknown.
Sure... but collateral damage is already a problem, as humans are very poor at actually hitting targets. People who are trained to shoot in high pressure situations are only in the area of 18-30% accurate.
I suspect if your goal was 30%< accuracy during an active shooting, it's absolutely within the realm of possibility today.
>According to a 2008 RAND Corporation study evaluating the New York Police Department’s firearm training, between 1998 and 2006, the average hit rate during gunfights was just 18 percent. When suspects did not return fire, police officers hit their targets 30 percent of the time.