Karen L. Myers and David N. Morley
For agent technology to be accepted in real-world applications, humans must be able to customize and control agent operations. One approach for providing such controllability is to enable a human supervisor to define guidance for agents in the form of policies that establish boundaries on agent behavior. This paper considers the problem of conflicting guidance for agents, making contributions in two areas: (a) outlining a space of conflict types, and (b) defining resolution methods that provide robustagent operation in the face of conflicts. These resolution methods combine a guidance-based preference relation over plan choices with an ability to extend the set of options considered by an agent when conflicts arise. The paper also describes a PRS-based guidance conflict-handling capability applied within a multiagent intelligence-gathering domain.