Modeling exceptions via commitment protocols

Abstract

This paper develops a model for exceptions and an approach for incorporating them in commitment protocols among autonomous agents. Modeling and handling exceptions is critical for successful applications of multiagent systems. Protocols help build multiagent systems, but traditional representations (such as finite state machines or Petri nets) inadequately model complex interactions and exceptions therein. Emerging commitment-based representations are promising, because they declaratively reflect the semantics of an interaction. However, current approaches lack a strong treatment of exceptions.This paper treats both expected and unexpected exceptions. A commitment protocol is modeled as a set of computations, each representing an allowed interaction and showing the evolving commitments of the participants. Exceptions are modeled via preference structures induced on these sets of computations. The preference structures statically show how expected exceptions are handled whereas the structures must be enhanced dynamically to handle unexpected exceptions. Our approach includes operators for composing protocols and exception handlers, whereby appropriate exception handlers can be dynamically introduced into a protocol as needed.The main contributions of this paper are (1) a framework for modeling and handling exceptions intelligently in commitment protocols and (2) a demonstration of the benefits of commitment protocols over traditional formalisms in handling exceptions.