Abstract
Controlling swarm dynamics is challenging and has long been an attractive research field because swarms provide a fundamental insight of locally interacting systems’ emergent behaviors. For example, a sheepdog type navigation control has been studied recently, where swarms consist of two different agents: passive sheep and active sheepdogs. In this paper, we focused on the swarm predator system with a swarm that has a number of passive agents and a single active predator agent. Recently, reservoir computing (RC) was introduced as a new way to control swarms. RC offers an easy and analyzable way to find optimal controllers. In this paper, we suggest a new way to read the swarms’ state for controlling swarm predator systems, named relatively ordered state (ROS), where the agents’ IDs are reordered at each time step by relative distances from the predator. The ROS is robust against the swarm’s initial condition’s difference, despite the simpleness and naturalness of the process of the ROS. We found that a swarm within the critical phases of order disorder phase transition like structures can bring out the swarm’s potential to be a reservoir both in open-loop and closed-loop experiments. In this closed-loop control, the predator determines its own movement via the collective dynamics of the swarm like a serpent eating its own tail in the classic “Ouroboros.”