Streaming Multi-agent Pathfinding

Authors: Mingkai Tang, Lu Gan, Kaichen Zhang

IJCAI 2025 | Venue PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental Experimental results indicate that ASCBS surpasses traditional MAPF solvers in terms of runtime for scenarios with prolonged working hours. This section presents experiments conducted to evaluate the computational efficiency and solution quality of the ASCBS on three grid-like graphs selected from the MAPF Benchmark Set [Stern et al., 2019], with varying scaling and features.
Researcher Affiliation Academia 1Hong Kong University of Science and Technology 2Hong Kong University of Science and Technology (Guangzhou) EMAIL
Pseudocode Yes The pseudocode of the high-level solver is presented in Algorithm 1. Algorithm 1 ASCBShigh
Open Source Code Yes The code is publicly available at https://github.com/tangmingkai/S-MAPF.
Open Datasets Yes This section presents experiments conducted to evaluate the computational efficiency and solution quality of the ASCBS on three grid-like graphs selected from the MAPF Benchmark Set [Stern et al., 2019], with varying scaling and features. The three chosen graphs are empty-8-8 (size: 8 8), random-64-64-10 (size: 64 64), and Paris 1 256 (size: 256 256).
Dataset Splits No For every scenario, 4 instances are generated with a randomly assigned initial start time for each agent stream within the range of [0, c 1] for the setting of circle time c and agent number n, resulting in a total of 100 instances for each setting.
Hardware Specification Yes All experiments are performed on a Ubuntu 20.04 computer equipped with an Intel Core i7-8700 CPU running at 3.2 GHz with 32GB of main memory.
Software Dependencies No All experiments are performed on a Ubuntu 20.04 computer equipped with an Intel Core i7-8700 CPU running at 3.2 GHz with 32GB of main memory.
Experiment Setup Yes We assessed the computational efficiency of the different implementations of ASCBS on the instances with varying numbers of agent streams, where the circle time is set at 3. Our evaluation includes instances with 10 agent streams and circle time 3.