Learning General Completable Reactive Plans

Melinda T. Gervasio

This paper presents an explanation-based learning strategy for learning general plans for use in an integrated approach to planning. The integrated approach augments a classical planner with the ability to defer achievable goals, thus preserving the construction of provably-correct plans while gaining the ability to utilize runtime information in planning. Proving achievability is shown to be possible without having to determine the actions to achieve the associated goals. A learning strategy called contingent explanation-based learning uses conjectured variables to represent the eventual values of plan parameters with unknown values a priori, and completers to determine these values during execution. An implemented system demonstrates the use of contingent EBL in learning a general completable reactive plan for spaceship acceleration.

This page is copyrighted by AAAI. All rights reserved. Your use of this site constitutes acceptance of all of AAAI's terms and conditions and privacy policy.