Abstract

Coordinating agents in dynamic and uncertain environments remains a fundamental challenge in multi-agent systems (MAS) research, particularly in contexts where the composition of agent organizations directly affects overall system performance. While significant effort has focused on task allocation and individual agent planning, predicting the systemic impact of organizational changes and selecting optimal organizational structures under uncertainty remain less explored in MAS. This paper addresses this challenge by introducing a decision-making framework that models structural reorganization as a Markov Decision Process (MDP), where actions represent organizational structures rather than individual agent behaviors, and organizational selection is guided by the anticipated impact on the overall system state. The proposed model captures the stochastic dynamics of multi-agent intervention and diverse agent capabilities through a probabilistic transition function, while a reward function guides the selection of coalition structures that maximize operational effectiveness. The framework is solved using value iteration and evaluated on the RoboCup Rescue simulation platform. Results show that the derived optimal policy identifies, at each decision step, an appropriate coalition structure that reduces system degradation while efficiently utilizing available agents.

Keywords