Introducing Preferences into Commitment Protocols

Abstract

Commitment protocols enable flexibility in agent interactions by utilizing the semantics of commitments to develop succinct declarative specifications for protocols that allow a large number of executions. As a consequence, commitment protocols enable agents to accommodate varying local policies and respond to exceptions. A consequent weakness of such protocols is that commitment protocols thus fail to distinguish between possible executions that are normal and those that may be allowed but are not ideal. This paper develops an approach for specifying preferences among executions that are allowed by a protocol. It captures sets of executions via a simple language and gives them a denotational characterization based on branching-time models. It shows how to incorporate the specifications into rulesets, thereby giving the specifications a natural operational characterization. The rulesets embed into a recent practical framework for protocols called OWL-P. The paper shows that the operational and denotational characterizations coincide.

We thank Amit Chopra, Nirmit Desai, and the anonymous referees for valuable comments. This research was supported partly by the NSF under grant DST-0139037 and partly by a DARPA project.