Communication is never perfect
A lot of multi-robot coordination theory implicitly assumes that every robot can talk to every other robot, constantly and reliably. Real deployments break that assumption in several ways. Robots often have limited communication range, so a robot on one side of a large warehouse or field may simply be out of reach of one on the other side. Bandwidth is limited too, so streaming every robot's full state to every other robot, or to a central coordinator, doesn't scale as the fleet grows. And communication links can drop out entirely, from radio interference, physical obstruction, or hardware failure.
This is a major reason decentralized, local-information approaches are attractive in practice, not just in theory. A coordination strategy that only requires a robot to communicate with its immediate neighbors keeps working even when the fleet as a whole has patchy, unreliable connectivity, whereas a strategy that assumes constant global communication can fail badly the moment that assumption breaks, exactly when robustness matters most.