Spatial Intelligence from a Cognitive Map Perspective:
A Survey

📚 Overview

Overview of spatial intelligence from a cognitive map perspective.

This survey revisits Spatial Intelligence from the perspective of Cognitive Maps, treating cognitive maps as the representational blueprint that connects spatial perception, reasoning, and generation. Under this view, diverse research directions can be understood through a shared question: how an internal spatial representation is constructed, maintained, reasoned over, and realized.

We organize the literature into three cognitive-map-centric processes:

• Perception: Construction of Cognitive Maps — how agents build internal spatial representations from local, partial, and multimodal observations.
• Reasoning: Inference with Cognitive Maps — how cognitive maps are used as embeddings, prompts, or APIs to support spatial inference and decision-making.
• Generation: Realization of Cognitive Maps — how internal maps are externalized into 3D scenes and dynamic world simulations.

This repository provides a curated list of papers analyzed in the survey, following the taxonomy proposed in Spatial Intelligence from a Cognitive Map Perspective: A Survey.

📖 Table of Contents

• 1. Perception: Construction of Cognitive Maps

  • 1.1 Metric Representation
  • 1.2 Relational Representation
  • 1.3 Hybrid Representation

• 2. Reasoning: Inference with Cognitive Maps

  • 2.1 Map as Embedding
  • 2.2 Map as Prompt
  • 2.3 Map as API

• 3. Generation: Realization of Cognitive Maps

  • 3.1 Static Scene Synthesis
  • 3.2 Dynamic World Simulation

• 4. Application Domains

  • 4.1 Open-Loop Spatial Cognition
  • 4.2 Closed-Loop Spatial Interaction

---

1. Perception: Construction of Cognitive Maps

1.1 Metric Representation

1.1.1 Explicit Geometry-based

2D Planar-based

3D Geometry-based

1.1.2 Parametric Coordinate-based

1.2 Relational Representation

1.2.1 Structured Graph-based

1.2.2 Serialized Graph-based

1.3 Hybrid Representation

1.3.1 Hierarchical Architecture

1.3.2 Feature Fusion

---

2. Reasoning: Inference with Cognitive Maps

2.1 Map as Embedding

2.1.1 Structured State Propagation

2.1.2 Latent Feature Matching

2.2 Map as Prompt

2.2.1 Textual Prompting

2.2.2 Visual Prompting

2.2.3 Multimodal Prompting

2.3 Map as API

2.3.1 Real-time State Snapshot

2.3.2 Persistent Spatial Memory

---

3. Generation: Realization of Cognitive Maps

3.1 Static Scene Synthesis

3.1.1 Map-based Retrieval

3.1.2 Map-to-Scene Generation

3.2 Dynamic World Simulation

---

4. Application Domains

4.1 Open-Loop Spatial Cognition

4.1.1 Spatial Question Answering

4.1.2 Indoor Scene Synthesis

4.1.3 Open-ended World Generation

4.2 Closed-Loop Spatial Interaction

4.2.1 Embodied Navigation

4.2.2 Embodied Manipulation

---

📌 Citation

If you find this survey or repository useful for your research, please consider citing:

@article{tian2026spatial,
  title={Spatial Intelligence from a Cognitive Map Perspective: A Survey},
  author={Tian, Yuxuan and Ji, Yuheng and Zheng, Xiaolong and Qin, Ziheng and Wang, Yipu and Zheng, Xinyi and Liu, Yuyang and Bai, Shuanghao and Li, Zhe and Wang, Liang and others},
  year={2026},
  publisher={Preprints}
}