Graph Attention-Guided Search for Dense Multi-Agent Pathfinding

TL;DR

This paper introduces LaGAT, a hybrid approach combining neural heuristics and search algorithms to efficiently solve dense multi-agent pathfinding problems in real-time, outperforming existing methods.

Contribution

The paper presents LaGAT, a novel hybrid framework integrating a graph attention-based neural heuristic with search algorithms for dense MAPF, improving performance over prior learning or search-only methods.

Findings

LaGAT outperforms purely search-based methods in dense scenarios.

LaGAT surpasses purely learning-based approaches in multi-agent pathfinding.

The hybrid approach effectively handles deadlocks and complex coordination tasks.

Abstract

Finding near-optimal solutions for dense multi-agent pathfinding (MAPF) problems in real-time remains challenging even for state-of-the-art planners. To this end, we develop a hybrid framework that integrates a learned heuristic derived from MAGAT, a neural MAPF policy with a graph attention scheme, into a leading search-based algorithm, LaCAM. While prior work has explored learning-guided search in MAPF, such methods have historically underperformed. In contrast, our approach, termed LaGAT, outperforms both purely search-based and purely learning-based methods in dense scenarios. This is achieved through an enhanced MAGAT architecture, a pre-train-then-fine-tune strategy on maps of interest, and a deadlock detection scheme to account for imperfect neural guidance. Our results demonstrate that, when carefully designed, hybrid search offers a powerful solution for tightly coupled, challenging multi-agent coordination problems.