Attention Focusing and Division of Labour in Swarm Warfare

In future conflicts, it is highly likely that autonomous machine collectives, also known as “swarms”, will engage each other without direct human intervention or supervision. Instead, control over these robotic combatants will most likely be indirect, taking the form of communicating strategic objectives, specifying the geographical limits of the area of operations, setting the rules of engagement etc., before deployment. One plausible scenario in the near to medium-term future is the use of an autonomous force of Unmanned Aerial Systems (UAS) to defend sensitive borders in remote, sparsely populated areas (e.g., the arctic region) against any incursion by a hostile swarm or other airborne threat. In this context, the development of distributed autonomous decision-making algorithms capable of improving the performance of the defending swarm without human guidance, e.g., in successfully and efficiently intercepting an enemy force, seems paramount. This specifically requires solving the problem of allocating proportional resources to engage multiple and potentially heterogeneous threats and to approximate optimal division of labour without central coordination. In this paper, we propose and evaluate such an algorithm, putting special emphasis on the influence of individual units having to make decisions based on incomplete information and using unreliable communication channels. Multiple scenarios are tested through Monte Carlo simulation, involving variable balance of power (relative swarm sizes), detection and communication ranges, and intruder strategy (invading “en masse” vs. repeated probing). The results are quantitatively analysed, allowing for the identification of important thresholds.