Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Large language model empowered participatory urban planning

Zhilun Zhou, Yuming Lin, Yong Li

TL;DR

The paper tackles the inefficiency and inclusivity gaps in participatory urban planning by embedding large-language-model agents into a three-module workflow: role-playing, collaborative generation, and feedback iteration. It implements CP, SP, and resident agents to generate and iteratively refine land-use plans across a community-scale set of plots, evaluated in two Beijing communities. Results show the LLM-driven approach delivers high satisfaction and inclusion, outperforms several baselines on need-aware metrics, and remains competitive with state-of-the-art RL on need-agnostic metrics, while offering transparent, natural-language rationales. This work demonstrates a scalable, low-cost augmentation to traditional planning that can enhance citizen participation and planning robustness, with potential for integration into real-world processes and broader stakeholder participation.

Abstract

Participatory urban planning is the mainstream of modern urban planning and involves the active engagement of different stakeholders. However, the traditional participatory paradigm encounters challenges in time and manpower, while the generative planning tools fail to provide adjustable and inclusive solutions. This research introduces an innovative urban planning approach integrating Large Language Models (LLMs) within the participatory process. The framework, based on the crafted LLM agent, consists of role-play, collaborative generation, and feedback iteration, solving a community-level land-use task catering to 1000 distinct interests. Empirical experiments in diverse urban communities exhibit LLM's adaptability and effectiveness across varied planning scenarios. The results were evaluated on four metrics, surpassing human experts in satisfaction and inclusion, and rivaling state-of-the-art reinforcement learning methods in service and ecology. Further analysis shows the advantage of LLM agents in providing adjustable and inclusive solutions with natural language reasoning and strong scalability. While implementing the recent advancements in emulating human behavior for planning, this work envisions both planners and citizens benefiting from low-cost, efficient LLM agents, which is crucial for enhancing participation and realizing participatory urban planning.

Large language model empowered participatory urban planning

TL;DR

The paper tackles the inefficiency and inclusivity gaps in participatory urban planning by embedding large-language-model agents into a three-module workflow: role-playing, collaborative generation, and feedback iteration. It implements CP, SP, and resident agents to generate and iteratively refine land-use plans across a community-scale set of plots, evaluated in two Beijing communities. Results show the LLM-driven approach delivers high satisfaction and inclusion, outperforms several baselines on need-aware metrics, and remains competitive with state-of-the-art RL on need-agnostic metrics, while offering transparent, natural-language rationales. This work demonstrates a scalable, low-cost augmentation to traditional planning that can enhance citizen participation and planning robustness, with potential for integration into real-world processes and broader stakeholder participation.

Abstract

Participatory urban planning is the mainstream of modern urban planning and involves the active engagement of different stakeholders. However, the traditional participatory paradigm encounters challenges in time and manpower, while the generative planning tools fail to provide adjustable and inclusive solutions. This research introduces an innovative urban planning approach integrating Large Language Models (LLMs) within the participatory process. The framework, based on the crafted LLM agent, consists of role-play, collaborative generation, and feedback iteration, solving a community-level land-use task catering to 1000 distinct interests. Empirical experiments in diverse urban communities exhibit LLM's adaptability and effectiveness across varied planning scenarios. The results were evaluated on four metrics, surpassing human experts in satisfaction and inclusion, and rivaling state-of-the-art reinforcement learning methods in service and ecology. Further analysis shows the advantage of LLM agents in providing adjustable and inclusive solutions with natural language reasoning and strong scalability. While implementing the recent advancements in emulating human behavior for planning, this work envisions both planners and citizens benefiting from low-cost, efficient LLM agents, which is crucial for enhancing participation and realizing participatory urban planning.
Paper Structure (20 sections, 8 equations, 7 figures, 2 tables)

This paper contains 20 sections, 8 equations, 7 figures, 2 tables.

Table of Contents

  1. Introduction
  2. Related work
  3. Participatory planning
  4. Multi-agent collaboration with large language models
  5. Methods
  6. LLM and the overall framework
  7. Role-playing
  8. Collaborative generation
  9. Feedback iteration
  10. Output
  11. Experiment
  12. Study area
  13. Initial conditions and planning requirements
  14. Evaluation metric
  15. Baseline
  16. ...and 5 more sections

Figures (7)

  • Figure 1: The framework of our proposed participatory planning method.
  • Figure 2: The satellite image and area division of two communities, Huilongguan (HLG) and Dahongmen (DHM).
  • Figure 3: Dialogue of asking residents' needs of facilities. Different residents have different needs corresponding to their profiles and special backgrounds.
  • Figure 4: An example dialogue of a sub-community planner discussing with residents and revising the plan.
  • Figure 5: The modification of the planning after revising by each sub-community planner and the reasons given by planners. Change are marked in bold solid lines.
  • ...and 2 more figures