Decision-theoretic Control of Failure Recovery

A numberof strategies exist for the recovery from execution-timeplan failures. Onedimensionin whichthese strategies differ is the degree of dependenceon the reliability andavailability of the planner'sknowledge.The best strategy, however,maybe dependenton a numberof considerations, including the type of plan failure, the critically of thefailure, the availabilityof resources,andthe reliability and availability of the specific knowledge involvedin a given plan failure instance. In our workon multi-agent,muRi-modal(centralized,distributedandlocal) planning,weare interested in intelligently selecting and applyingfailure recoverystrategies that are appropriateto the situation and that can copewith uncertainty. Weare examininga decision-theoretic approachto diagnoseplan failures and to dynamicallyselect frommultiple failure recoverystrategies whenan execution-timeplan failure Occurs.

• Type: Conference
• Year: 1994
• Where: AIPS
• Authors: Lisa J. Burnell
• Keywords: AIPS 1994, Artificial Intelligence, Execution-timeplan Failure, Plan Failure, Plan Failure Instance