MONAD: A Flexible Architecture for Multi-Agent Control

Research in multi-agent systems has led to the development of many multi-agent control architectures. However, we believe that there is currently no known optimal structure for multi-agent control since the effectiveness of any particular architecture varies depending on the domain of the problem. Therefore, deployment of multi-agent teams would be significantly sped up by a development and deployment environment which would allow designers to easily modify the architecture. In this paper, we present a flexible team-oriented programming and execution architecture, MONAD, which integrates hierarchical behaviorbased control, multi-agent coordination mechanisms, and agenttask allocation services. MONAD uses a novel scripting language that allows designers to easily modify the team structure, behavior hierarchy, applicability conditions, and arbitration methods, in pursuit of the best solution for a particular problem. We have evaluated the MONAD architecture within a well-accepted adversarial game environment, GameBots, to enable qualitative comparison of different control techniques. In this environment, we were able to rapidly design and test several teams of agents who used role, preference, or a combination of role and preference arbitration and observed that these different teams varied in their performance characteristics.