Authors:
Fatemeh Nargesian
and
Gholamreza Ghassem-Sani
Affiliation:
Sharif University of Technology, Iran, Islamic Republic of
Keyword(s):
AI Planning, hierarchical task network planning, machine learning, decision tree, domain knowledge.
Related
Ontology
Subjects/Areas/Topics:
Artificial Intelligence
;
Formal Methods
;
Informatics in Control, Automation and Robotics
;
Intelligent Control Systems and Optimization
;
Planning and Scheduling
;
Simulation and Modeling
;
Symbolic Systems
Abstract:
In this paper, we describe LHTNDT, an algorithm that learns the preconditions of HTN methods by examining plan traces produced by another planner. LHTNDT extracts conditions for applying methods by using decision tree based algorithm. It considers the state of relevant domain objects in both current and goal states. Redundant training samples are removed using graph isomorphism. Our experiments, LHTNDT converged. It can learn most of preconditions correctly and quickly. 80% of our test problems were solved by preconditions extracted by ¾ of plan traces needed for full convergence.