Notice: The reproducibility variables underlying each score are classified using an automated LLM-based pipeline, validated against a manually labeled dataset. LLM-based classification introduces uncertainty and potential bias; scores should be interpreted as estimates. Full accuracy metrics and methodology are described in [1].

HTN Planning as Heuristic Progression Search

Authors: Daniel Höller, Pascal Bercher, Gregor Behnke, Susanne Biundo

JAIR 2020 | Venue PDF | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental Our empirical evaluation shows that the resulting system outperforms the state of the art in search-based HTN planning.
Researcher Affiliation Academia Daniel H oller EMAIL Pascal Bercher EMAIL Gregor Behnke EMAIL Susanne Biundo EMAIL Institute of Artificial Intelligence, Ulm University, Ulm, Germany
Pseudocode Yes Algorithm 1: The canonical progression-based HTN planning algorithm as grounded, fringe-based formulation as given by H oller et al. (2018a). Algorithm 2: Optimized algorithm presented by H oller et al. (2018a). Algorithm 3: Novel progression algorithm.
Open Source Code Yes We call our new system PANDApro (indicating the progression search). The source code of the PANDA planners and all domains and problems are available online at www.uni-ulm.de/in/ki/panda.
Open Datasets Yes The UM-Translog, Satellite, and Woodworking domains are HTN versions of the corresponding domains known from classical planning. They are further described by Bercher et al. (2014). Smartphone A domain describing the task of operating a smartphone. A description is also given by Bercher et al. (2014). The source code of the PANDA planners and all domains and problems are available online at www.uni-ulm.de/in/ki/panda.
Dataset Splits No The paper lists several domains used for evaluation (UM-Translog, Satellite, Woodworking, Smart Phone, Rover, Transport, Entertainment, PCP) and states they are available online with the source code. However, it does not specify any training/test/validation splits for these problem instances; it treats them as a set of problems to be solved.
Hardware Specification Yes All experiments ran on Xeon E5-2660 v3 CPUs with a base frequency of 2.60 GHz, a memory limit of 4 GB and time limit of 10 minutes.
Software Dependencies No The paper mentions that the search engine was reimplemented in C++ and uses components like PANDA for preprocessing. It also refers to classical heuristics such as hadd, h FF, and h LM -Cut by their respective papers. However, it does not provide specific version numbers for the C++ compiler, PANDA, or any other critical software libraries used in their implementation.
Experiment Setup Yes All experiments ran on Xeon E5-2660 v3 CPUs with a base frequency of 2.60 GHz, a memory limit of 4 GB and time limit of 10 minutes. We included 27 heuristic search configurations of our system: {Alg. 1, Alg. 2, Alg. 3} {rcadd, rc FF, rc LM-Cut} {G, A , WA }.