Previously having discussed humanoid robot design and large scale operations, we can now discuss what happens when the combat robots are separated from each other.
This requires that the command structure has a realistic idea of how the war is going. Historically speaking, while knowing fairly accurately does not guarantee victory, not knowing (either deliberately or accidentally) results in a rather messy war. Whatever heuristic you're using to attempt to predict your future success in a war will most likely be complex, but for now let's assume it is realistic and relatively accurate.
The hypothesis presented here relates to gambling. When one is winning, one should do the regular thing to maintain their advantage. When one is losing (especially by a lot), one should engage in high risk/high reward behaviours, lest one be squashed by their opponent's superior position. This flow, from low risk/low reward to risky, is actually quite a common theme in game playing. It does not guarantee victory (especially if one is losing from the start), but almost always results in the better option regardless of your adversary's choices. This typically works in a zero-sum game, as warfare often is (diplomatic solutions are probably not where you are using your combots).
In such a case where your robot army is believed to be winning (often through superiority in numbers and position), if a group or a single robot is cut off from the main body and cannot communicate, they should either retreat to a known body of more robots, or defend their current position if they believe they are being reinforced and that position is valuable enough to be worth a defence of some casualties.
In a case where your robot army is believed to be losing, the army will start taking riskier and riskier actions as it loses more and more. While it will not guarantee victory, it can prolong such a war to the point where an opposing army won't necessarily want to fight any more, thus perhaps preventing total defeat. Whereas fighting traditionally and retreating to masses of connected robots will result in a fairly ordinary war of attrition, where failure is all but guaranteed.
If you follow this line of thinking to extremes, this results in guerilla warfare, tying down a far larger force of opposing soldiers than an army ever could. See France circa 1942 and Afghanistan whenever.
With an army of robots, this ebb and flow from traditional warfare to fourth generation warfare is actually quite natural. Certainly superior and more realistic than the current bone-headed sticking to traditional warfare that results in a crapton of casualties.
Also, it makes robots in losing armies act like Rambo, which is horrifying, hilarious, and kind of cool.