Limiting disruption in multiagent replanning

Abstract

Multiagent systems sometimes undergo changes that cause coordination commitments to become insufficient or out of date, such that the coordinated agent plans need to be repaired or replaced. When recoordination becomes necessary, disruption to the commitments made by the agents in their original plans should be minimized. We approach the problem of minimizing disruption by augmenting pre-existing coordination technology by developing metrics and automated processes for it to rank and potentially recommend new coordination commitments more rapidly. In this paper, we explain, examine, and evaluate our new metrics and processes, demonstrating empirically that flexible measures of disruption can streamline the coordination process.