Cover Page Verification of Behavior in Multi-Agent Systems

Effective planning and scheduling in multi-agent systems (MAS) requires the development of representations that can model their unique problem-solving requirements. This paper analyzes both the macro-behavior (behavior of a MAS w.r.t. system goals) and the micro-behavior (behavior of individual agents w.r.t. to local goals) of MAS. The Sensible Agent architecture provides a representation of the agent organizations in which the respective agent participates to pursue and accomplish goals. Specifically, this autonomy level representation defines an agent’s interactions with other agents in the system and, thus, the macro-behavior of agent problem solving organizations in the system. The Sensible Agent structure supports plan generation and execution of the resulting micro-behaviors. This research also extends traditional satisfiability (SAT) planners to model agent-level microbehavior for planning. The extensions use Binary Decision Diagrams and formal verification toolkits to improve the forecasting ability of agent planners. The resulting planner uses Computation Tree Logic (CTL) formulas to check local stability while retaining the speed and expressiveness of traditional satisfiability planners. The combination of microand macro-behavior modeling techniques greatly increases the control and stability of MAS.