부산대학교 도서관

Swarm cohesion is a prime contributor to the success of swarm guidance. However, a limited sensing range of the agents could lead to fragmentation of the swarm, which cascades to mission failure. Shepherding is a swarm guidance approach that relies on a finite number of agents to control a proportionally larger number of swarm members. We propose a connectivity-aware shepherding mechanism for improving the cohesion of the swarm in the presence of sensing constraints. A graph-based model of the flock is integrated with a particle swarm optimisation algorithm and a data clustering algorithm to identify the most appropriate herding point for influencing the swarm to move and to guide them towards the goal. The path to this herding point is optimised by particle swarm optimisation. The proposed methodology is evaluated with multiple swarm configurations. A comparison between the proposed methodology and classic shepherding demonstrates the superior performance of the former, which is demonstrated by a reduction of task completion time to half and a doubling in mission success rate.