Cascaded control of multi-agent systems

Intelligent software agents are an important enabling technology for the next generation of distributed information systems. As agent-based applications spread and enter the realm of survivable and embedded systems, the issues of control become prominent in the development of agent architectures. This paper discusses the problem of multi-level control in distributed agent systems and presents the control approach and infrastructure elements of the open source agent architecture "Cougaar". We motivate its design from a control theory perspective, discuss the characteristics of an agent system as the controlled process, and present an example of agent infrastructure-level control.