[Paper Review] Large Language Model for Participatory Urban Planning

Zhilun Zhou, Yuming Lin+2 more|arXiv (Cornell University)|Feb 27, 2024

Geographic Information Systems Studies11 citations

TL;DR

This paper uses a multi-agent LLM framework to simulate planner and resident roles for participatory urban planning, employing fishbowl discussions to iteratively revise a land-use plan and optimize for resident satisfaction and inclusion.

ABSTRACT

Participatory urban planning is the mainstream of modern urban planning that involves the active engagement of residents. However, the traditional participatory paradigm requires experienced planning experts and is often time-consuming and costly. Fortunately, the emerging Large Language Models (LLMs) have shown considerable ability to simulate human-like agents, which can be used to emulate the participatory process easily. In this work, we introduce an LLM-based multi-agent collaboration framework for participatory urban planning, which can generate land-use plans for urban regions considering the diverse needs of residents. Specifically, we construct LLM agents to simulate a planner and thousands of residents with diverse profiles and backgrounds. We first ask the planner to carry out an initial land-use plan. To deal with the different facilities needs of residents, we initiate a discussion among the residents in each community about the plan, where residents provide feedback based on their profiles. Furthermore, to improve the efficiency of discussion, we adopt a fishbowl discussion mechanism, where part of the residents discuss and the rest of them act as listeners in each round. Finally, we let the planner modify the plan based on residents' feedback. We deploy our method on two real-world regions in Beijing. Experiments show that our method achieves state-of-the-art performance in residents satisfaction and inclusion metrics, and also outperforms human experts in terms of service accessibility and ecology metrics.

Motivation & Objective

• Motivate participatory urban planning as a time- and cost-efficient alternative to traditional methods.
• Introduce LLM-based planner and resident agents to model diverse resident needs.
• Develop a fishbowl discussion mechanism to scale discussions in large communities.
• Enable planner revision of initial land-use plans based on resident feedback.

Proposed method

• Assign planner and resident roles to LLM agents via prompts, including region maps and profiles.
• Generate an initial land-use plan for a region; simulate resident discussions within communities.
• Use a fishbowl discussion mechanism to alternatingly include inner-circle speakers and outer-circle listeners.
• Summarize discussion history to maintain manageable context and update the plan accordingly.
• Partition regions into communities and run multiple discussion rounds to balance needs.
• Evaluate against baselines using need-agnostic (Service, Ecology) and need-aware (Satisfaction, Inclusion) metrics.

Experimental results

Research questions

• RQ1 Can LLM-based agents simulate a planner-resident participatory planning process effectively?
• RQ2 Do resident discussions (including a fishbowl mechanism) improve alignment with diverse resident needs?
• RQ3 How does the LLM-based approach compare to baselines and human designers on service, ecology, satisfaction, and inclusion?
• RQ4 What is the impact of discussion rounds on planning quality and efficiency?

Key findings

• The proposed method achieves the best performance on need-aware metrics (Satisfaction and Inclusion) across both datasets (HLG and DHM).
• On HLG, Satisfaction reaches 0.787 and Inclusion 0.773, outperforming several baselines; DRL excels in need-agnostic metrics but underperforms in need-aware metrics.
• In need-agnostic metrics, the method is generally second-best, often surpassing human experts in Service and Ecology.
• Ablation shows both role-play and discussion contribute to performance; removing either reduces need-aware metrics by 4.7–8.0% and 4.1–6.3%, respectively.
• Increasing fishbowl rounds from 1 to 3 improves Satisfaction and Inclusion noticeably, with diminishing returns beyond 3 rounds.
• The approach scales discussions to large resident populations via the fishbowl mechanism while maintaining reasonable ecology trade-offs.