Real-Time Heuristic Search with LTLf Goals
Authors: Jaime Middleton, Rodrigo Toro Icarte, Jorge Baier
IJCAI 2022 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | In our experimental evaluation over standard benchmarks we show LTL-LRTA*A may outperform LTL-LRTA* substantially for a variety of LTLf goals. |
| Researcher Affiliation | Collaboration | Jaime Middleton1, 4 , Rodrigo Toro Icarte1,2 , Jorge A. Baier1,2,3 1Department of Computer Science, Pontificia Universidad Cat olica de Chile, Santiago, Chile 2Centro Nacional de Inteligencia Artificial CENIA, Santiago, Chile 3Instituto Milenio Fundamentos de los Datos, Santiago, Chile 4Tucar Sp A, Santiago, Chile jamiddleton@uc.cl, rodrigo.toro@ing.puc.cl, jabaier@ing.puc.cl |
| Pseudocode | Yes | Algorithm 1: LTL-LRTA*, a simple variant of LSSLRTA* that solves an LSP by carrying out search over the cross-product representation. |
| Open Source Code | Yes | Our source code and appendix are publicly available at https://github.com/Jamidd/RTHS-with-LTLf-Goals |
| Open Datasets | Yes | For our experiments we use standard grids, Starcraft maps and mazes, using a number of LTLf goals. We ran experiments on three families of grid problems proposed by Sturtevant [2012]: Rooms, Starcraft, and Maze. |
| Dataset Splits | No | The paper mentions running experiments on benchmarks but does not specify explicit training, validation, or test dataset splits (e.g., 80/10/10 split, or specific sample counts for each partition). |
| Hardware Specification | No | The paper does not provide specific details about the hardware used for experiments, such as CPU/GPU models, memory, or specific computing environments. |
| Software Dependencies | No | The paper does not list specific versions for any software dependencies, libraries, or frameworks used for the experiments. |
| Experiment Setup | Yes | In total, we ran 27, 000 experiments. These considered 5 LTLf goals, 3 domains (i.e., rooms, starcraft, and maze), 10 maps per domain, 5 problem instances per map (where each instance is a different placement of the letters in the map), 6 lookahead values (with k {32, 64, 128, 256, 512, 1024}), and 3 heuristics (h1 φ, h M φ , and h M φ .). When placing letters in the maps, we either placed three letters of each type (i.e., three a s, three b s, etc) or twenty-five. |