A Survey of Large Language Model-Powered Spatial Intelligence Across Scales: Advances in Embodied Agents, Smart Cities, and Earth Science

TL;DR

The paper addresses how large language models can equip spatial intelligence across embodied, urban, and Earth-scale domains, motivated by insights from human spatial cognition. It introduces a unifying taxonomy and framework that connect spatial memory, knowledge, and abstract reasoning in LLMs to practical applications ranging from robotic navigation to GIS-assisted planning and climate geoscience. By synthesizing literature across disciplines, the authors highlight key advances, representative systems, and emergent patterns, while identifying core challenges in representation, evaluation, data integration, and interpretability. The work emphasizes the potential of cross-domain, multi-scale spatial intelligence to inform future AI systems and real-world decision-making, and it points toward world-model integration and human-in-the-loop approaches as central avenues for progress.

Abstract

Over the past year, the development of large language models (LLMs) has brought spatial intelligence into focus, with much attention on vision-based embodied intelligence. However, spatial intelligence spans a broader range of disciplines and scales, from navigation and urban planning to remote sensing and earth science. What are the differences and connections between spatial intelligence across these fields? In this paper, we first review human spatial cognition and its implications for spatial intelligence in LLMs. We then examine spatial memory, knowledge representations, and abstract reasoning in LLMs, highlighting their roles and connections. Finally, we analyze spatial intelligence across scales -- from embodied to urban and global levels -- following a framework that progresses from spatial memory and understanding to spatial reasoning and intelligence. Through this survey, we aim to provide insights into interdisciplinary spatial intelligence research and inspire future studies.

Figures (7)

• Figure 1: Multiple scale spatial intelligence in real world: from embodied spatial intelligence to earth spatial intelligence.
• Figure 2: A taxonomy of large language model-empowered spatial intelligence with representative examples.
• Figure 3: This figure illustrates the core concepts of Spatial Memory and Knowledge in LLMs. LLMs build their spatial memory and knowledge from both internal and external sources to perform tasks like question answering, navigation, and geolocalization, while also facing challenges such as hallucination mitigation and knowledge editing.
• Figure 4: Conceptual framework of Abstract Spatial Reasoning. The framework illustrates three primary dimensions of spatial reasoning capabilities: qualitative reasoning, geometric reasoning, and graph reasoning. LLMs still face the challenge of bridging language understand to abstract spatial cognition.
• Figure 5: A simple schematic of embodied spatial intelligence. The framework illustrates two sequential stages: spatial perception and understanding and spatial interaction and navigation.