Malte Helmert and Jendrik Seipp receive ICAPS Awards
The Research Group Artificial Intelligence from the Department of Mathematics and Computer Science received two ICAPS Awards this year.
The ICAPS Influential Paper Award went to Prof. Dr. Malte Helmert and Prof. Dr. Carmel Domshlak. Traditionally, domain-independent planning heuristics were based on one of four ideas: delete relaxations, critical paths, abstractions, and landmarks. In their paper “Landmarks, Critical Paths and Abstractions: What's the Difference Anyway?” the researchers examined new connections between existing planning heuristics and obtained a new heuristic from their analysis. This "Landmark Cut" heuristic is not only very interesting from a theoretical point of view, but also overshadowed all previously available planning techniques for optimal classic planning. The award is given to two papers each year. To qualify for the award, papers must be published at least 10 years before the year of the current conference.
The ICAPS Best Dissertation Award went to Dr. Jendrik Seipp, who in his dissertation "Counterexample-Guided Cartesian Abstraction Refinement and Saturated Cost Partitioning for Optimal Classical Planning" presents a new family of permitted heuristics for classic planning. These heuristics are based on Cartesian abstractions, which are derived by means of counter-example-controlled abstraction simplifications. Different methods are introduced to create several diversified abstractions, since a single abstraction often does not provide enough information for complex planning tasks. In order to combine the information of the different abstractions in an admissible manner, Jendrik Seipp presents a new algorithm called "Saturated Cost Partitioning". It considers the heuristics sequentially and uses the minimum amount of costs that preserves all heuristic estimates for the current heuristic before passing the remaining costs to subsequent heuristics until all heuristics have been served this way. Since "Saturated Cost Partitioning" heuristics are very informative and very efficient to calculate using abstraction heuristics, they represent the state-of-the-art in optimal classic planning.
The official award ceremony will take place during the 30th International Conference on Automated Planning and Scheduling in Nancy, France. Due to the current situation, it was postponed to October 26-30, 2020.
Picture of Nancy, France on the DMI homepage: descubriendoelmundo @flickr.