The Motivation for Dynamic Decision-Making Frameworks in Multi-Agent Systems
K. Suzanne Barber, Cheryl English Martin · Series in machine perception and artificial intelligence · 2001
The coordination of agent actions in a multi-agent system requires establishing an organizational policy (sometimes implicit) that specifies the locus of decision-making control and the authority of decision-makers to assign tasks. For this research, the distribution of decision-making control and authority-over relationships among agents for a particular goal or set of goals is called a decision-making framework. Most multi-agent systems maintain static decision-making frameworks that range from hierarchically controlled to local, distributed control. The research presented here proposes that decision-making frameworks should instead be dynamically adapted to run-time conditions. Experiments extending the authors ’ previous work show that no one decision-making framework performs best across various situations that may be faced at run-time. In fact, which decision-making framework works best varies greatly across situations, where “best ” is defined based on four performance measures. These experiments provide a clear motivation for the implementation of dynamic decision-making frameworks in multi-agent systems.