A hierarchical model of agent based on skill, rules, and knowledge

摘要

The principal aim of this research is to structure agents according to different situations that have to deal with. To achieve this, we propose a conceptual model with a hierarchical structure defined by the skill-rule-knowledge (S-R-K) levels. At the skill level, the agent deals with routines and her behavior is governed by stored patterns of predefined procedures, that map directly an observation (i.e. perception) to an action. The rule-based level represents more conscious behavior and it deals with familiar situations. This behavior is generally conventionally described by a set of heuristics. Finally, the knowledge-based level accounts for unfamiliar situations for which know-how or rules are not available. An implementation of this model has been done in the context of a multiagent environment to confirm our mean expectation: the perceptual processing (i.e., S and R levels) is fast, effortless and is propitious for coordinatedactiween agents, whereas the analytical problem solving (i.e., K level) is slow, laborious and can lead to conflicts between agents.