Bilateral Spatial Reasoning about Street Networks: Graph-based RAG with Qualitative Spatial Representations

TL;DR

This paper enhances LLMs' ability to generate pedestrian route instructions using qualitative spatial relations within street networks, aiming to improve spatial reasoning and navigation guidance.

Contribution

It introduces a graph-based RAG approach incorporating qualitative spatial representations to improve LLMs' spatial reasoning for pedestrian navigation.

Findings

Improved accuracy in route instruction generation.

Enhanced spatial reasoning capabilities in LLMs.

Effective use of qualitative spatial relations in street networks.

Abstract

This paper deals with improving the capabilities of Large Language Models (LLM) to provide route instructions for pedestrian wayfinders by means of qualitative spatial relations.