Learning from goal interactions in planning: goal stack analysis and generalization

Authors:
Kwang Ryel Ryu;Keki B. Irani
Affiliations:
Artificial Intelligence Laboratory, Department of Electrical Engineering and Computer Science, The University of Michigan, Ann Arbor, MI;Artificial Intelligence Laboratory, Department of Electrical Engineering and Computer Science, The University of Michigan, Ann Arbor, MI
Venue:
AAAI'92 Proceedings of the tenth national conference on Artificial intelligence
Year:
1992

Citing 8
Cited 0

Planning for conjunctive goals

Artificial Intelligence
Deriving subgoal ordering constraints prior to problem solving and planning

Deriving subgoal ordering constraints prior to problem solving and planning
A structural theory of explanation-based learning

A structural theory of explanation-based learning
Learning search control knowledge for planning with conjunctive goals

Learning search control knowledge for planning with conjunctive goals
Learning Search Control Knowledge: An Explanation-Based Approach

Learning Search Control Knowledge: An Explanation-Based Approach
A Computer Model of Skill Acquisition

A Computer Model of Skill Acquisition
Ordering problem subgoals

IJCAI'89 Proceedings of the 11th international joint conference on Artificial intelligence - Volume 2
The role of preprocessing in problem solving systems: "An ounce of reflection is worth a pound of backtracking"

IJCAI'77 Proceedings of the 5th international joint conference on Artificial intelligence - Volume 1

Quantified Score

Hi-index	0.00

Visualization

Abstract

This paper presents a methodology which enables the derivation of goal ordering rules from the analysis of problem failures. We examine all the planning actions that lead to failures. If there are restrictions imposed by a problem state on taking possible actions, the restrictions manifest themselves in the form of a restricted set of possible bindings. Our method makes use of this observation to derive general control rules which are guaranteed to be correct. The overhead involved in learning is low because our method examines only the goal stacks retrieved from the leaf nodes of a failure search tree rather than the whole tree. Empirical tests show that the rules derived by our system PAL, after sufficient training, performs as well as or better than those derived by systems such as PRODIGY/EBL and STATIC.