combat-capable autonomous robots
does not necessarily imply "human-level intelligence capable of dealing with all of the situations that may arise once the shooting has stopped". Human environments and human interactions are complex, and fully understanding them would need AGI and if you have human-equivalent AI with guns then All Bets Are Off anyway, so lets discard that idea for the moment.
High quality, low latency audio and video links over long distances are surprisingly awkward in the real world without having to operate in a war zone on a near-airless radiation-soaked rock where you can't necessarily rely on infrastructure to be there or functional, and in the absense of a troposphere or ionosphere and in the face of the possibility of cheap antisatellite rockets (low G! thin air! scifi space tech!) and ECM, radio links are always going to be unreliable and slow. Your scout bots may be supervised but the supervisors can't get a good handle on what is actually happening.
Robots go in first, but humans are needed to mop up and secure the area and handle prisoners and the wounded and captured equipment and work out exactly what happened and who was involved, because the robot supervisors may have little if any idea.
