Online Bridged Pruning for Real-Time Search with Arbitrary Lookaheads

Authors: Carlos Hernandez, Adi Botea, Jorge A. Baier, Vadim Bulitko

IJCAI 2017 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental An experimental evaluation shows that our pruning often improves the performance of a base real-time search algorithm by over an order of magnitude. This allows our implemented system to outperform state-of-the-art real-time search algorithms used in the evaluation. 7 Empirical Evaluation
Researcher Affiliation Collaboration Chilean Center for Semantic Web Research IBM Research, Ireland Pontificia Universidad Cat olica de Chile Universidad Andr es Bello, Chile University of Alberta, Canada
Pseudocode Yes Algorithm 1: Real-time Heuristic Search Framework; Algorithm 2: PALMA; Algorithm 3: PALMACONNECT
Open Source Code No No explicit statement about open-source code release or a link to a repository for the described methodology.
Open Datasets Yes We took maps from the commonly used Moving AI pathfinding repository [Sturtevant, 2012].
Dataset Splits No For each map we generated 100 random solvable problems.
Hardware Specification Yes Experiments are run on a 2.60GHz Intel Core i7 machine under Linux, as a single-threaded process.
Software Dependencies No We implemented all algorithms in C over a common code base.
Experiment Setup Yes For both algorithms we use the best parameters as reported in the original publications (w = 8 for w LSS-LRTA* and C = 8 for EDA*). We implemented all algorithms in C over a common code base. For LSS-LRTA* and its variants, we use a binary heap for Open in which ties are broken in favor of larger g-values. We took maps from the commonly used Moving AI pathfinding repository [Sturtevant, 2012]. We used all five 1024 × 1024 Star Craft maps, 30 maze maps of size 512 × 512 (corridor widths of 4, 8, 16), all 512 × 512 maps World of Warcraft III (WC3) and all room maps of size 512 × 512. Finally, as a fifth benchmark, we consider the same maze maps, but with 1% of their obstacles removed. For each map we generated 100 random solvable problems. Each map is modeled as an undirected eight-connected grid, with costs set to 1 for cardinal moves and √2 for diagonal moves. The octile distance is used as a heuristic. By varying the lookahead value (i.e., number of expanded states per episode) LSS (1, 10, 100 and 200).