Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Generative AI Meets Future Cities: Towards an Era of Autonomous Urban Intelligence

Dongjie Wang, Chang-Tien Lu, Xinyue Ye, Tan Yigitcanlar, Yanjie Fu

TL;DR

This paper advocates a deep generative AI view of urban planning, reframing land-use configuration as a data-driven generation problem conditioned on geospatial contexts, mobility patterns, and human instructions. It surveys three core models—LUCGAN (GAN-based), CLUVAE (VAE-based), and IHPlanner (Transformer-based)—each with distinct representations, objectives, and limitations, and proposes a generic representation-generation framework to unify these approaches. The authors emphasize knowledge-guided, fairness-aware, and human-in-the-loop planning, arguing for hybrid architectures that combine the strengths of generative, probabilistic, and sequential models. The envisioned future highlights automated data preparation, informative representations (e.g., spatial-attributed graphs), and fairness-aware planning, calling for simulations, agentic AI, and interdisciplinary collaboration to translate AI-driven planning into real-world, equitable urban outcomes.

Abstract

The two fields of urban planning and artificial intelligence (AI) arose and developed separately. However, there is now cross-pollination and increasing interest in both fields to benefit from the advances of the other. In the present paper, we introduce the importance of urban planning from the sustainability, living, economic, disaster, and environmental perspectives. We review the fundamental concepts of urban planning and relate these concepts to crucial open problems of machine learning, including adversarial learning, generative neural networks, deep encoder-decoder networks, conversational AI, and geospatial and temporal machine learning, thereby assaying how AI can contribute to modern urban planning. Thus, a central problem is automated land-use configuration, which is formulated as the generation of land uses and building configuration for a target area from surrounding geospatial, human mobility, social media, environment, and economic activities. Finally, we delineate some implications of AI for urban planning and propose key research areas at the intersection of both topics.

Generative AI Meets Future Cities: Towards an Era of Autonomous Urban Intelligence

TL;DR

This paper advocates a deep generative AI view of urban planning, reframing land-use configuration as a data-driven generation problem conditioned on geospatial contexts, mobility patterns, and human instructions. It surveys three core models—LUCGAN (GAN-based), CLUVAE (VAE-based), and IHPlanner (Transformer-based)—each with distinct representations, objectives, and limitations, and proposes a generic representation-generation framework to unify these approaches. The authors emphasize knowledge-guided, fairness-aware, and human-in-the-loop planning, arguing for hybrid architectures that combine the strengths of generative, probabilistic, and sequential models. The envisioned future highlights automated data preparation, informative representations (e.g., spatial-attributed graphs), and fairness-aware planning, calling for simulations, agentic AI, and interdisciplinary collaboration to translate AI-driven planning into real-world, equitable urban outcomes.

Abstract

The two fields of urban planning and artificial intelligence (AI) arose and developed separately. However, there is now cross-pollination and increasing interest in both fields to benefit from the advances of the other. In the present paper, we introduce the importance of urban planning from the sustainability, living, economic, disaster, and environmental perspectives. We review the fundamental concepts of urban planning and relate these concepts to crucial open problems of machine learning, including adversarial learning, generative neural networks, deep encoder-decoder networks, conversational AI, and geospatial and temporal machine learning, thereby assaying how AI can contribute to modern urban planning. Thus, a central problem is automated land-use configuration, which is formulated as the generation of land uses and building configuration for a target area from surrounding geospatial, human mobility, social media, environment, and economic activities. Finally, we delineate some implications of AI for urban planning and propose key research areas at the intersection of both topics.
Paper Structure (35 sections, 8 figures)

This paper contains 35 sections, 8 figures.

Table of Contents

  1. When Urban Planning Meets Artificial Intelligence
  2. Deep Generative Urban Planning: Analogy, Terminologies, and Algorithmic Framework
  3. A Generative AI Perspective: Automated Urban Planning as Deep Generative Learning
  4. Key Terminologies in Deep Generative Urban Planning
  5. Target Area
  6. Geospatial Contexts
  7. Urban Functional Zone
  8. Land-use Configuration
  9. Human Instruction
  10. A Generic Framework of Deep Generative Urban Planning
  11. Representation: Perceiving Geospatial Forms, Human Mobility, Social Interactions, and Human Planner Instructions
  12. Generation: Learning to Generate Land-use Configurations
  13. Representative Models of Deep Generative Urban Planning
  14. Automated Urban Planner via Generative Adversarial Learning
  15. Research Gap
  16. ...and 20 more sections

Figures (8)

  • Figure 1: An AI-based urban planner generates optimal land-use configurations by taking into account realistic planning constraints and the input of human urban experts.
  • Figure 2: An analogy between urban planning and deep generative AI.
  • Figure 3: Illustration of the target area and geospatial contexts. (a) Geospatial contexts encircle the target area from different directions. (b) The spatial attributed graph contains all features of geospatial contexts.
  • Figure 4: Urban functional zone and land-use configuration. (a) Zone-level planning is a 2-D matrix, which provides high-level guidance for grid-level planning. (b) Grid-level planning is represented by a 3-D tensor where we reserve the 3rd dimension for POI as each grid may contain multiple POI categories.
  • Figure 5: An overview of the LUCGAN framework: This model comprises two primary components. Firstly, the generator endeavors to create optimal land-use configurations by utilizing the embedding of the surrounding environment. Secondly, the discriminator's objective is to assign elevated scores to real favorable configurations, while attributing lower scores to both unfavorable real and generated configurations.
  • ...and 3 more figures