Handling non-local dead-ends in Agent Planning Programs
Authors: Lukas Chrpa, Nir Lipovetzky, Sebastian Sardina
IJCAI 2017 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | We provide experimental evidence suggesting the technique yields significant benefits. |
| Researcher Affiliation | Academia | Luk aˇs Chrpa Charles University in Prague & Czech Technical University Prague, Czech Republic chrpaluk@fel.cvut.cz, Nir Lipovetzky University of Melbourne Melbourne, Australia nir.lipovetzky@unimelb.edu.au, Sebastian Sardina RMIT University Melbourne, Australia sebastian.sardina@rmit.edu.au |
| Pseudocode | Yes | Algorithm 1 Online APP realization with dead-end filtering. |
| Open Source Code | No | The paper mentions using and integrating with the LAPKT toolkit [Ramirez et al., 2015], which has a URL provided in the references. However, it does not state that the authors' *own* code for the described methodology is being released or is publicly available. |
| Open Datasets | Yes | To evaluate our proposal, we designed experiments using six planning domains from the International Planning Competition. The domains mentioned are Airport, Floortile, Glued Blocks World, Matching Blocks World, Logistics, and Woodworking. |
| Dataset Splits | No | The paper states 'For each domain, we have generated 20 APPs' and describes how they were used for evaluation, but it does not specify any training, validation, or test splits for these APPs or the underlying data within them. |
| Hardware Specification | Yes | All the experiments were run on i7-7700 3.6 Ghz CPU with 16GB of RAM. |
| Software Dependencies | No | The paper names specific planners like LPG, Lm-cut, and DFS+, and the LAPKT toolkit, along with their respective citations, but it does not provide specific version numbers for these software components. |
| Experiment Setup | Yes | The maxsteps parameter was set to infinity for acyclic APPs (the realization is successful if a leaf program state is reached), i.e, for Airport, Floortile, Glued -Bw and Woodworking, and to 10 for APPs with cycles, i.e., Logistics and Matching-Bw, in order to assure that every cycle has been performed at least twice. Since LPG is a randomized planner and thus with different runs solution plans might differ even for the same planning problem, we performed 20 runs for each APP. |