Knowledge-based Behavior Specification

Many critical multi-agent applications require flexibility in both, individual and collective agent behavior that cannot be addressed by existing models. An example could be an agent system destined for unexpected situation handling and recovery. The paper proposes a framework to model unexpected and exceptional situations relying on goal-oriented situated planning viewpoint integrating both knowledge-based and behavior-based ideas. The paper introduces an ontology-based knowledge model, Scenario Knowledge Base framework which represents experts' knowledge about agent capabilities to act as well as a syntactic behavior patterns structure given over the set of actions the agents are capable to execute. This framework is supported by reusable software core intended for an online situated planning that uses the former to dynamically infer all possible courses of actions of agents of a multi-agent system to achieve ultimate goal starting from current agent states.