Cohesion: Keeping Independently-Moving Agents Close Together

Abstract

Many algorithms for groups of moving physical agents require that the agents remain close enough together to coordinate cooperation. Most such algorithms do not provide facilities for maintaining inter-agent cohesion. Previous cohesion algorithms generally either supersede other mobility objectives (flocking) or constrain an entire collective to act as a single entity (formations). We present a framework for defining and maintaining agent cohesion without otherwise restricting agent behaviors. Our cohesion framework is designed as a drop-in filter on agent behaviors, preventing cohesion- breaking maneuvers without specifying particular behaviors for individual agents. The cohesion framework has similarity to many collision avoidance algorithms with respect to its applicability in conjunction with task- specific maneuvering algorithms. We prove that our framework guarantees cohesion and provide approximation techniques that ensure it is computationally tractable for broad classes of agents. We demonstrate the versatility of our approach by example.