A Generalization of Automated Planning Using Dynamically Estimated Action Models. Weiss, E. In Proceedings of the 20th ICAPS Doctoral Consortium (ICAPS DC 2022), pages 1–3, 2022.
A Generalization of Automated Planning Using Dynamically Estimated Action Models [pdf]Paper  A Generalization of Automated Planning Using Dynamically Estimated Action Models [link]Presentation  abstract   bibtex   9 downloads  
Representing real-world planning problems is a major open subject. Standard planning modeling languages are fully declarative, making it challenging to use them for expressing complex mathematical functions, that are often required for describing the effects of actions. Recent approaches turn to external sources of information, such as simulators or black-box modules, to overcome such modeling limitations. This paper proposes a novel approach to represent and solve planning problems, by starting with partial declarative action models and incrementally refining them during planning by invoking domain-specific external modules. Since these might be computationally expensive, we provide the planner the ability to trade-off modeling uncertainty against computation time, to meet target plan accuracy. Results that were obtained for planning with dynamic estimation of action costs are sketched, and planned work, together with open challenges, are further detailed.

Downloads: 9