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

2025-12-17

AI Summary

Large language models (LLMs) struggle to accurately comprehend and generate qualitative spatial relations—such as “turn left,” “adjacent to,” or “cross”—in pedestrian wayfinding tasks. To address this, we propose a novel framework integrating qualitative spatial reasoning with graph-structured retrieval-augmented generation (RAG). Our approach innovatively embeds the RCC-8 and OPRA qualitative spatial calculi into street-network topological graphs, enabling semantic-aligned bidirectional spatial reasoning. Coupled with graph neural networks, it supports structured modeling of spatial relations and context-aware generation. Evaluated on real-world urban pedestrian navigation benchmarks, our method achieves a 27% improvement in instruction accuracy and reduces spatial relation errors by 41% over state-of-the-art LLM baselines. This work establishes a new paradigm for interpretable and robust route guidance in embodied navigation, bridging symbolic spatial reasoning with neural language generation.

Technology Category

• Planning, Routing, and Scheduling: Planning with Language Models
• Knowledge Representation and Reasoning: Geometric, Spatial, and Temporal Reasoning
• Reasoning under Uncertainty: Relational Probabilistic Models

Application Category

• Semantics and Knowledge: Methods to enhance, augment, integrate or synergize semantic models such as knowledge graphs and LLMs
• Graph Algorithms and Modeling for the Web: Foundation models and LLMs for Web-related graphs
• Search and Retrieval-Augmented AI: Large language models for search

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.

Problem

• Enhances LLMs for pedestrian route instructions
• Uses qualitative spatial relations for navigation
• Applies graph-based RAG to street networks

Innovation

• Graph-based RAG with qualitative spatial representations
• Bilateral spatial reasoning about street networks
• Improving LLM capabilities for pedestrian route instructions