Abstract
Multi-agent systems often require the formation of subgroups of varying sizes to solve their assigned task. In dynamic environments, it becomes an online problem as groups must continuously adjust their size and composition to maintain overall efficiency and stability. We propose a decentralized group regulation framework inspired by natural fission-fusion dynamics. Our approach combines a gossip-based group size estimation algorithm with a probabilistic behavioral control model. We enable anonymous agents to asynchronously estimate their group sizes using only local information and to autonomously execute group merging or splitting behaviors. Simulation results show stable and adaptive group size regulation under different communication modes and dynamic conditions. Our unified mechanism regulates both the size and number of groups and has potential to serve as a theoretical and algorithmic foundation for this fundamental but underexplored problem.
About the speaker
Tianfu Zhang is a Ph.D. candidate in the Cyber-Physical Systems Group at the University of Konstanz.