Enhancing PIBT via Multi-Action Operations

TL;DR

The paper tackles online Lifelong MAPF with rotation constraints by extending PIBT to Enhanced PIBT (EPIBT), introducing multi-action operations, revisiting, and inheritance to maintain speed while handling time-consuming rotations. EPIBT is integrated with Large Neighborhood Search (LNS) and Graph Guidance (GG) to achieve state-of-the-art throughput in the LMAPF-T setting, outperforming PIBT, Causal PIBT, and other baselines across standard maps. Theoretical guarantees akin to PIBT are maintained under a simple-cycle graph assumption, with a concrete bound showing fast progression for the highest-priority agent and a formal complexity bound. Empirical results demonstrate strong improvements in both online LMAPF-T and LMAPF, with EPIBT+LNS+GG delivering top performance and robust heatmap-based flow characteristics, indicating practical impact for large-scale, rotation-aware multi-robot routing.

Abstract

PIBT is a rule-based Multi-Agent Path Finding (MAPF) solver, widely used as a low-level planner or action sampler in many state-of-the-art approaches. Its primary advantage lies in its exceptional speed, enabling action selection for thousands of agents within milliseconds by considering only the immediate next timestep. However, this short-horizon design leads to poor performance in scenarios where agents have orientation and must perform time-consuming rotation actions. In this work, we present an enhanced version of PIBT that addresses this limitation by incorporating multi-action operations. We detail the modifications introduced to improve PIBT's performance while preserving its hallmark efficiency. Furthermore, we demonstrate how our method, when combined with graph-guidance technique and large neighborhood search optimization, achieves state-of-the-art performance in the online LMAPF-T setting.

Figures (17)

• Figure 1: Spider plot demonstrating the relative performance of the evaluated approaches. Solid lines represent PIBT-like approaches without additional components, dotted lines indicate methods utilizing LNS, and dashed lines denote approaches that also incorporate GG. Related solvers are indicated by the same color.
• Figure 2: Cells and states reachable with different operation length. Different cell colors indicate the minimum required operation length to reach the corresponding cell. Arrows and numbers near them indicate the actual state and number of actions required to reach it.
• Figure 3: Visualization of maps used for the empirical evaluation.
• Figure 4: Ablation study of different enhancements incorporated into EPIBT and its evaluation with different operation lengths.
• Figure 5: Comparison of EPIBT+LNS+GG with operation lengths 3, 4, and 5 with both winners of LoRR competition and Causal PIBT+traffic flow on the online LMAPF-T setting.