On Succinct Groundings of HTN Planning Problems
Authors: Gregor Behnke, Daniel Höller, Alexander Schmid, Pascal Bercher, Susanne Biundo9775-9784
AAAI 2020 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | In order to ascertain the quality of the grounding found by panda PI and its runtime performance when doing so, we have conducted an empirical evaluation. We have compared panda PI against three other grounders for HTN planning problems: PANDA s grounder, GTOHP s grounder (Ramoul et al. 2017), and Tree-Rex s grounder (Schreiber et al. 2019). A per-instance comparison of the runtime between panda PI, PANDA, GTOHP, and Tree-Rex is shown in Fig. 5. A statistical summary of timings and grounding sizes is shown in Tab. 2. |
| Researcher Affiliation | Academia | Gregor Behnke, Daniel H oller, Alexander Schmid, Pascal Bercher, Susanne Biundo Institute of Artificial Intelligence, Ulm University, D-89069 Ulm, Germany {gregor.behnke, daniel.hoeller, alexander-1.schmid, susanne.biundo}@uni-ulm.de, pascal.bercher@alumni.uni-ulm.de Pascal Bercher is now at the College of Engineering and Computer Science, the Australian National University |
| Pseudocode | Yes | Algorithm 1 shows the overall mechanism of PANDA s grounder. ... Algorithm 1: PANDA s grounding procedure Overview. Variables are: Pg grounded primitive actions, Ag grounded abstract tasks, Mg grounded methods |
| Open Source Code | Yes | The code of panda PI is publicly available at https://github. com/galvusdamor/panda PIgrounder. |
| Open Datasets | Yes | The benchmark set comprises 330 problem instances from 17 domains. ... All instance are available for download at pandapi.hierarchical-task.net/domains. |
| Dataset Splits | No | The paper mentions evaluating on a benchmark set and how different grounders were tested on subsets (e.g., totally-ordered instances), but it does not specify train/validation/test splits, percentages, or sample counts needed to reproduce data partitioning for typical model training and evaluation. |
| Hardware Specification | Yes | All experiments were conducted on an Intel Xeon E5-2660 with a 20 GB RAM limit for each instance. |
| Software Dependencies | No | The paper mentions that 'panda PI is written in C++, while PANDA is written in Scala' and compares its implementation to 'Fast Downward s grounder (Helmert 2009)', but it does not provide specific version numbers for these or any other software dependencies. |
| Experiment Setup | No | The paper mentions general experimental conditions such as 'a 20 GB RAM limit' and 'no time-limit', and notes that some grounders only handle 'totally-ordered HTN planning problems'. However, it does not provide specific hyperparameter values, model initialization details, or other system-level configuration settings that would typically constitute a detailed experimental setup. |