Multi-Agent Motion Planning for Differential Drive Robots Through Stationary State Search

Authors: Jingtian Yan, Jiaoyang Li

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

Reproducibility Variable Result LLM Response
Research Type Experimental Empirically, we tested our methods on the single-shot grid map domain and the lifelong warehouse domain. Our method shows up to 400% improvements in terms of throughput compared to existing methods.
Researcher Affiliation Academia Jingtian Yan, Jiaoyang Li Carnegie Mellon University EMAIL, EMAIL
Pseudocode Yes Algorithm 1: Stationary SIPP (SSIPP) Algorithm 2: Partial Stationary Expansion Algorithm 3: Pseudocode for Windowed-SSIPP
Open Source Code Yes The source code for our method is publicly accessible at https://github.com/JingtianYan/MASS-AAAI.
Open Datasets Yes We evaluated all methods on four-neighbor grid maps, including empty (empty-32-32, size: 32 32), random (random-32-32-10, size: 32 32), room (room-64-64-8, size: 64 64), den520d (den520d, size: 256 257), Boston (Boston 0 256, size: 256 256), warehouse-small (warehouse-10-20-10-2-1, size: 161 63), warehouse-large (warehouse-20-40-10-22, size: 340 164) from the Moving AI benchmark (Stern et al. 2019), and the sortation-center map (size: 500 140) from the LMAPF Competition (Chan et al. 2024).
Dataset Splits No For each map, we conducted experiments with a progressive increment in the number of agents, using the 25 random scenarios from the benchmark set. For lifelong MAMPD, the paper states: "We run each method for 1,000 simulation time seconds and average the results over 7 runs." The paper does not provide specific train/test/validation dataset splits but rather uses predefined scenarios/instances or simulation time and runs.
Hardware Specification Yes We conducted all experiments on an Ubuntu 20.04 machine equipped with an AMD 3990x processor and 188 GB of memory. Our code was executed using a single core for all computations.
Software Dependencies No We implemented both our and baseline methods in C++. We conducted all experiments on an Ubuntu 20.04 machine. The paper mentions the programming language (C++) and operating system (Ubuntu 20.04) but does not provide specific version numbers for any key software libraries or dependencies used.
Experiment Setup Yes All agents adhered to the same kinodynamic constraints, where the speed is bounded by the range of [0, 2] cell/s, while the acceleration is confined to [ 0.5, 0.5] cell/s2. Each agent has a speed limit from [0, 2] m/s and an acceleration limit from [ 0.5, 0.5] m/s. Table 2 also lists 'tw' values (20 s, 25 s, 30 s, 40 s) for the lifelong MAMPD evaluation.