Multi-Goal Multi-Agent Path Finding via Decoupled and Integrated Goal Vertex Ordering

Authors: Pavel Surynek12409-12417

AAAI 2021 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental We evaluated HCBS and SMT-HCBS on standard benchmarks from movingai.com (Sturtevant 2012). Representative part of the results is presented in this section. and Runtime Results We divided the tests into three categories with respect to the size of maps. Results for small instances derived from the empty-16-16 map are shown in Figure 3. Three different cases with the number of goal vertices per agent: 1 (corresponds to MAPF), 2, 4, and 8 are tested.
Researcher Affiliation Academia Pavel Surynek Czech Technical University in Prague, Faculty of Information Technology, Th akurova 9, 160 00 Praha 6, Czechia pavel.surynek@fit.cvut.cz
Pseudocode Yes Algorithm 1: HCBS algorithm for MG-MAPF. and Algorithm 2: SMT-based MG-MAPF solver
Open Source Code Yes For the full reproducibility of the presented results we provide a complete source code of our solvers and detailed experimental data on author s web: http://users.fit.cvut.cz/ surynpav/research and in git repository: https://github.com/surynek.
Open Datasets Yes We evaluated HCBS and SMT-HCBS on standard benchmarks from movingai.com (Sturtevant 2012).
Dataset Splits No The paper mentions generating instances and evaluating them, but does not specify explicit training, validation, or test dataset splits (e.g., percentages, sample counts, or predefined citations) for reproducibility.
Hardware Specification Yes All experiments were run on system consisting of 200 Xeon 2.8 GHz cores, 1TB RAM, running Ubuntu Linux 18.
Software Dependencies Yes HCBS and SMT-HCBS were implemented in C++. The SMT-HCBS solver is built on top of the Glucose 3.0 SAT solver (Audemard and Simon 2018)
Experiment Setup Yes The experimental evaluation has been done on diverse instances consisting of 4-connected grid maps ranging in sizes from small to large. Random MAPF scenarios from movingai.com are used to generate MG-MAPF instances. To obtain instances of various difficulties we varied the number of agents while the number of goal vertices per agent was set as constant. As defined in the benchmark set, 25 different instances are generated per number of agents. and The success rate is the ratio of the number of instances out of 25 per number of agents that the solver managed to solver under the time limit of 5 minutes.