[Paper Review] Generative AI Meets Future Cities: Towards an Era of Autonomous Urban Intelligence

Dongjie Wang, Chang‐Tien Lu+3 more|arXiv (Cornell University)|Apr 8, 2023

Human Mobility and Location-Based Analysis22 citations

TL;DR

The paper surveys and develops deep generative AI frameworks (LUCGAN, CLUVAE, IHPlanner) to automate land-use configuration generation using geospatial, mobility, social data and human guidance.

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.

Motivation & Objective

• Motivate the integration of urban planning with AI to enhance sustainability, livability, and resilience.
• Formulate automated land-use configuration as a deep generative learning task.
• Synthesize existing generative model approaches for urban planning (GANs, VAEs, Transformers) with domain knowledge.
• Highlight challenges, limitations, and future research directions at the AI–urban planning interface.

Proposed method

• Frame urban planning as image-like generation of land-use configurations conditioned on geospatial contexts and human instructions.
• Propose LUCGAN: embedding surrounding contexts via a spatial graph and generating configurations with an extended GAN.
• Propose CLUVAE: a conditional variational encoder–decoder that uses human guidance and context to produce structured land-use outputs.
• Propose IHPlanner: a transformer-based hierarchical generator with a Functionalizer module and multi-attention to capture spatial dependencies.
• Discuss objective functions for generative models (ELBO for VAEs, adversarial loss for GANs, likelihoods for autoregressive/flow/energy-based models) and domain-guided regularization.
• Outline limitations and future directions for data sparsity, human-in-the-loop control, and fairness.

Experimental results

Research questions

• RQ1 How can land-use configurations be quantified and generated automatically from surrounding contexts and human instructions?
• RQ2 What architectures (GANs, VAEs, transformers) best support automated urban planning with domain knowledge and spatial hierarchies?
• RQ3 How can human guidance be integrated to produce diverse, robust, and fair land-use configurations?
• RQ4 What are the limitations of current automated planners in capturing planning semantics and expert requirements?
• RQ5 What is the role of spatial hierarchy and socio-economic interactions in generating plausible urban plans?

Key findings

• LUCGAN embeds surrounding contexts into a latitude–longitude–channel representation and uses an extended GAN to generate land-use configurations.
• CLUVAE introduces a conditional variational encoder–decoder that incorporates human guidance and spatial hierarchies to improve robustness and diversity.
• IHPlanner uses a transformer-based, hierarchical generation with a Functionalizer to encode human and environmental constraints and a multi-attention mechanism across subareas.
• The paper demonstrates qualitative and quantitative evaluation with real-world data, highlighting improvements and remaining challenges in achieving human-centric and spatially coherent plans.
• Limitations include the difficulty of encoding rich planning semantics and achieving stable, real-world performance, along with data sparsity and evaluation metric issues.