Data-Efficient Multi-Agent Spatial Planning with LLMs

2025-02-26

AI Summary

This work addresses multi-agent taxi dispatch in graph-structured road networks, aiming to achieve low-data-dependency and robust spatial decision-making—minimizing total passenger waiting time—by leveraging the world knowledge embedded in pre-trained large language models (LLMs). Method: We propose the first multi-agent planning framework integrating LLM zero-shot reasoning with one-at-a-time forward rollout, enhanced by semantic-aware prompt engineering, lightweight fine-tuning, and structured road-network modeling. Contribution/Results: Our method surpasses existing SOTA using only 1/50 of environment interactions. It achieves strong zero-shot performance, with prompt effectiveness markedly improved upon incorporating readily available contextual information. Crucially, the LLM adapts to dynamic environmental changes via pure natural-language prompts. The core contribution is the empirical validation that LLMs’ intrinsic spatial commonsense knowledge enables effective and generalizable coordination in multi-agent decision-making, even under sparse supervision and structural constraints.

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
• Graph Algorithms and Modeling for the Web: Foundation models and LLMs for Web-related graphs
• Search and Retrieval-Augmented AI: Search Tool Learning with LLM: Teaching LLMs to invoke search and make use of retrieved information

Abstract

In this project, our goal is to determine how to leverage the world-knowledge of pretrained large language models for efficient and robust learning in multiagent decision making. We examine this in a taxi routing and assignment problem where agents must decide how to best pick up passengers in order to minimize overall waiting time. While this problem is situated on a graphical road network, we show that with the proper prompting zero-shot performance is quite strong on this task. Furthermore, with limited fine-tuning along with the one-at-a-time rollout algorithm for look ahead, LLMs can out-compete existing approaches with 50 times fewer environmental interactions. We also explore the benefits of various linguistic prompting approaches and show that including certain easy-to-compute information in the prompt significantly improves performance. Finally, we highlight the LLM's built-in semantic understanding, showing its ability to adapt to environmental factors through simple prompts.

Problem

• Leverage LLMs for multi-agent decision making
• Optimize taxi routing to minimize waiting time
• Enhance performance with linguistic prompting techniques

Innovation

• Leverage pretrained LLMs for decision making
• Use zero-shot and fine-tuning techniques
• Incorporate linguistic prompts for efficiency