The TOAD System for Totally Ordered HTN Planning

Authors: Daniel Höller

JAIR 2024 | Venue PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental Our evaluation on the 2020 IPC benchmark set shows that our system is competitive with the state of the art in HTN planning. 8. Empirical Evaluation This section is divided into four parts. First we investigate the impact of the applied optimizations in Section 8.1, then we compare different search configurations of Toad in Section 8.2 and have a look at unsolved instances in Section 8.3. Lastly, we compare Toad to the state of the art in HTN planning in Section 8.4.
Researcher Affiliation Academia Daniel Höller EMAIL Saarland University Saarland Informatics Campus Saarbrücken, Germany
Pseudocode Yes Algorithm 1 Algorithm by Nederhof (2000a, Figure 2) to translate (non-self-embedding) context-free grammars to finite automata. 1 procedure make fa(q0, α, q1)
Open Source Code Yes Next we describe our implementation, which is available online5. 5. toad.hierarchical-task.net
Open Datasets Yes We evaluate Toad on the benchmark set of the 2020 International Planning Competition. Table 1: Properties of the IPC 2020 benchmark set.
Dataset Splits No The paper evaluates the Toad system on the benchmark set of the 2020 International Planning Competition, which consists of various planning problem instances. However, it does not mention specific training/test/validation dataset splits as typically found in machine learning experiments for reproducing data partitioning.
Hardware Specification Yes All experiments in this section ran on Intel Xeon E5-2650 CPUs with 2.30 GHz (one core per job), a time limit of 30 minutes, and a memory limit of 8 GB.
Software Dependencies No The paper mentions using "Fast Downward (Fd) system" and "Open Fst Library" but does not provide specific version numbers for these software components.
Experiment Setup Yes All experiments in this section ran on Intel Xeon E5-2650 CPUs with 2.30 GHz (one core per job), a time limit of 30 minutes, and a memory limit of 8 GB. To combine mq(dfad, ff ) and ehc(dfad), we configured Fd to do an iterated search, first performing EHC with hdfad followed by the multi-queue mq(dfad, ff )... The configuration with 2 minutes is used, which we denote i[ehc, mq] (for iterated search).