Lifelong Multi-Agent Path Finding in Large-Scale Warehouses

Authors: Jiaoyang Li, Andrew Tinka, Scott Kiesel, Joseph W. Durham, T. K. Satish Kumar, Sven Koenig11272-11281

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

Reproducibility Variable Result LLM Response
Research Type Experimental We empirically evaluate RHCR with a variety of MAPF solvers and show that it can produce high-quality solutions for up to 1,000 agents (= 38.9% of the empty cells on the map) for simulated warehouse instances, significantly outperforming existing work.
Researcher Affiliation Collaboration Jiaoyang Li,1 Andrew Tinka,2 Scott Kiesel,2 Joseph W. Durham,2 T. K. Satish Kumar,1 Sven Koenig1 1 University of Southern California 2 Amazon Robotics
Pseudocode Yes Algorithm 1: The low-level search for Windowed MAPF solvers generalizing Multi-Label A* (Grenouilleau, van Hoeve, and Hooker 2019).
Open Source Code Yes The code is available at https://github.com/Jiaoyang-Li/RHCR.
Open Datasets Yes We use the map in Figure 3a from (Liu et al. 2019). ... We use the map in Figure 3b.
Dataset Splits No The paper simulates environments and assigns tasks dynamically, but it does not specify explicit training, validation, or test dataset splits in the traditional sense.
Hardware Specification Yes We conduct all experiments on Amazon EC2 instances of type m4.xlarge with 16 GB memory.
Software Dependencies No The paper mentions software components like C++, SIPP, and SCIPP, and specific MAPF solvers (CBS, ECBS, CA*, PBS), but it does not provide version numbers for any of these dependencies.
Experiment Setup Yes For RHCR, we use time horizon w = 20 and replanning period h = 5. ... We simulate 5,000 timesteps for each experiment with potential function threshold p = 1. ... ECBS with suboptimality factor 1.1, CA* with random restarts.