LLMDR: LLM-Driven Deadlock Detection and Resolution in Multi-Agent Pathfinding

2025-03-02

AI Summary

To address frequent deadlocks, poor generalization, and limited scalability in multi-agent path finding (MAPF), this paper pioneers the integration of large language models (LLMs) into MAPF deadlock resolution. We propose a dynamic deadlock identification and resolution framework that synergistically combines LLMs with learning-based MAPF models (e.g., GNN- or RL-based). The framework unifies semantic understanding, symbolic reasoning, and strategy-oriented prompt engineering to enable both interpretable pattern recognition in deadlock graphs and generation of customizable, generalizable resolution policies. Evaluated on standard MAPF benchmarks, our method achieves a 92.4% success rate in deadlock scenarios—representing an absolute improvement of 18.7%—and scales effectively to hundreds of agents. The implementation is publicly available.

Technology Category

• Planning, Routing, and Scheduling: Planning with Language Models
• Multiagent Systems: Multiagent Planning
• Machine Learning: Large Multimodal Models (LMMs)

Application Category

• Semantics and Knowledge: Data modeling to support human-machine intelligence, including LLMs agents, intelligent system behavior, explanations, and user-friendly interactions
• User Modeling, Personalization and Recommendation: Large Language Models (LLM) for user modeling and recommendation
• Graph Algorithms and Modeling for the Web: Foundation models and LLMs for Web-related graphs

Abstract

Multi-Agent Pathfinding (MAPF) is a core challenge in multi-agent systems. Existing learning-based MAPF methods often struggle with scalability, particularly when addressing complex scenarios that are prone to deadlocks. To address these challenges, we introduce LLMDR (LLM-Driven Deadlock Detection and Resolution), an approach designed to resolve deadlocks and improve the performance of learnt MAPF models. LLMDR integrates the inference capabilities of large language models (LLMs) with learnt MAPF models and prioritized planning, enabling it to detect deadlocks and provide customized resolution strategies. We evaluate LLMDR on standard MAPF benchmark maps with varying agent numbers, measuring its performance when combined with several base models. The results demonstrate that LLMDR improves the performance of learnt MAPF models, particularly in deadlock-prone scenarios, with notable improvements in success rates. These findings show the potential of integrating LLMs to improve the scalability of learning-based MAPF methods. The source code for LLMDR is available at: https://github.com/ssbacc/llmdr-dhc

Problem

• Addresses scalability issues in Multi-Agent Pathfinding (MAPF).
• Resolves deadlocks in complex MAPF scenarios using LLMs.
• Improves performance of learning-based MAPF models.

Innovation

• Integrates LLMs with MAPF models
• Detects and resolves deadlocks effectively
• Improves success rates in complex scenarios