Abstract:
This thesis contributes to the research in regards to the provision of the
conceptual and architectural models to integrate the object-based HTN planning with
deductive database framework, a development of a novel encoding method as
systematic translation from OCLh domain to ASP planner, and using the HTN
knowledge of an object-based domain to prune problem instance or reduce branching
factor based on deductive database techniques.
In addition, we exploited two implementation models by using two of the most
efficient deductive database systems available to date, namely, DLV and Smodels. The
OCLh2DDB translation module was proposed and developed to generate the entire set
of our object-based DDB-HTN encoding. Moreover, our ASP planner is evaluated
based on the three benchmarks of zeno-travel, blocks world and rocket world, as well
as compared encoding aspects and time requirements against Trans(·).
Our method demonstrated slightly better performance for the first answer set
generation over Trans(·), and our small-sized ASP planner works better in the ordered
task decomposition as well as non-ordered manner. We, however, found that using
grounding process in the ASP systems reduced the efficiency in large instance
problems like zeno-travel domain. In comparison with the dedicated planner, HyHTN,
we obtain better performance and also identify some interesting aspects of our
approach when solving the less sophisticated domains in operator-based version.
The discussions stem from our endeavor in the exploitation of deductive
database expressive power that would benefit the knowledge-enriched and objectbased
aspects of HTN domains.