GuardPIBT: Counterfactually Gated Neural Guidance for Ultra-Large-Scale 3D Multi-Agent Path Finding

2026-09-28

AI Summary

This study addresses the congestion problem in large-scale 3D multi-agent path planning under dense traffic, where the PIBT algorithm suffers from insufficient single-step goal-directedness. We propose GuardPIBT, a framework that leverages graph attention networks and a counterfactual gating mechanism to perform residual re-ranking exclusively over PIBT candidate actions while preserving its validity checking and backtracking mechanisms. Furthermore, efficient scalability is achieved through swarm-adaptive grouping, asynchronous cached inference, and selective repair strategies. Experimental results demonstrate that the proposed method enables violation-free operation for up to one hundred thousand agents in both 2D and 3D environments, significantly improving the success rate and computational efficiency of ultra-large-scale path planning.

Abstract

Large-scale 3D multi-agent path finding becomes increasingly difficult under dense traffic. Priority Inheritance with Backtracking (PIBT) scales well, but its one-step goal-directed ordering may become insufficient under dense interactions and large-scale congestion. We present GuardPIBT, which augments rather than replaces the PIBT executor: neural predictions only propose residual reorderings of PIBT's native candidates, while final actions remain determined by PIBT. First, local graph attention models nearby interactions, while global source--goal transport features provide population-level coordination context for candidate reordering. Second, a counterfactual group gate filters reorderings whose closed-loop effects may degrade coordination. Third, for ultra-large populations, population-adaptive grouping preserves decision granularity, asynchronous cached inference amortizes neural computation, and selective repair resolves long-tail agents. PIBT retains validity checking, priority inheritance, and backtracking throughout. Experiments with up to 100,000 agents demonstrate reliable completion across 2D and 3D environments, including all three 100,000-agent warehouse runs with zero audited graph violations. The project website is available at {\color{magenta}\texttt{https://guardpibt.github.io/GuardPIBT/}}.

Problem

• Multi-Agent Path Finding
• 3D environments
• Large-scale congestion
• Dense traffic
• Priority Inheritance with Backtracking

Innovation

• Multi-Agent Path Finding
• Neural Guidance
• Counterfactual Gating
• Graph Attention
• Scalability