Abstrakt
Heuristic-guided forward search is currently the leading approach to automated domain-independent planning. Many types of heuristics have been devised in the past and successfully used on various planning problems.
These heuristics, however, often require some parameters to be set properly in order to achieve a good performance. As tuning these parameters automatically proved to be difficult, practitioners often require assistance of a human expert.
Current research in the area focuses of utilizing meta-heuristics or machine learning techniques for this task. In this paper, we present the pattern database heuristic and two state-of-the-art techniques for tuning its parameters automatically based on meta-heuristics.
We summarize the current methods and propose several ways for further improvement.