Multiagent plan repair by combined prefix and suffix reuse

摘要

Deterministic domain-independent multiagent planning is an approach to coordination of cooperative agents with joint goals. Provided that the agents act in an uncertain and dynamic environment, such plans can fail. The straightforward approach to recover from such situations is to compute a new plan from scratch, that is to replan. Even though, in a worst case, plan repair or plan re-use does not yield an advantage over replanning from scratch, there is a sound evidence from practical use that approaches trying to repair the failed original plan can outperform replanning in selected problems. One of the possible plan repairing techniques is based on preservation of fragments of the older plans. This work theoretically analyses complexity of plan repairing approaches based on preservation of fragments of the original plan and experimentally studies three practical aspects affecting its efficiency in various multiagent settings. We focus both on the computational, as well as the communication efficiency of plan repair in comparison to replanning from scratch and we report on the influence of the following properties on the efficiency of plan repair: (1) the number of involved agents in the plan repairing process, (2) inter-dependencies among the repaired actions, and finally (3) particular modes of re-use of the older plans.