Trace-Based Specification of Law and Guidance Policies for Multi-Agent Systems

摘要

Policies have traditionally been a way to specify properties of a system. In this paper, we show how policies can be applied to the Organization Model for Adaptive Computational Systems (OMACS). In OMACS, policies may constrain assignments of agents to roles, the structure of the goal mode! for the organization, or how an agent may play a particular role. In this paper, we focus on policies limiting system traces; this is done to leverage the work already done for specification and verification of properties in concurrent programs. We show how traditional policies can be characterized as law policies; that is, they must always be followed by a system. In the context of multiagent systems, law policies limit the flexibility of the system. Thus, in order to preserve the system flexibility while still being able to guide the system into preferring certain behaviors, we introduce the concept of guidance policies. These guidance policies need not always be followed; when the system cannot continue with the guidance policies, they may be suspended. We show how this can guide how the system achieves the top-level goal while not decreasing flexibility of the system. Guidance policies are formally defined and, since multiple guidance policies can introduce conflicts, a strategy for resolving conflicts is given.