LiPUP-MA: A Residential Experience-centric Multi-Agent Framework for Living-in-the-loop Participatory Urban Planning

Researchers introduce LiPUP-MA, an LLM-based multi-agent framework that reimagines participatory urban planning through iterative living simulations rather than static preference gathering. The system uses an experience bank and spatially-constrained planning agents to translate residential feedback into coherent urban design revisions, demonstrating improvements over traditional planning methodologies.

This research addresses a fundamental limitation in how AI assists urban planning: the assumption that stakeholder preferences can be captured in isolated, one-time consultations. LiPUP-MA introduces a cyclical approach where simulated residential experiences feed directly into plan refinement, creating a feedback loop that mirrors how actual city-dwellers interact with their environments over time.

The framework tackles two critical technical challenges that have constrained AI planning applications. First, it grounds abstract living experiences within specific urban contexts through a graph-based experience bank, preventing feedback from becoming generic or disconnected from spatial realities. Second, it bridges the gap between subjective resident feedback and actionable planning modifications through a spatially-constrained planner agent that harmonizes experiential data with visual and geospatial information.

For urban developers and municipal planners, this represents a meaningful advancement in democratic planning processes. Traditional participatory planning often struggles with scalability—engaging diverse stakeholders repeatedly becomes prohibitively expensive. LLM-based agents can simulate thousands of resident interactions at computational scale, surfacing conflicts and preferences that might otherwise remain hidden until implementation.

The research demonstrates that iterative LiPUP cycles produce measurably better urban plans by both conventional metrics and living-experience metrics. As cities worldwide grapple with housing shortages and community resistance to development, tools that legitimately incorporate diverse residential perspectives while maintaining planning coherence gain practical importance. Future implementations could integrate real behavioral data from existing neighborhoods to enhance simulation fidelity and applicability to actual urban contexts.

• →LiPUP-MA uses iterative living simulations to transform static urban planning into a dynamic, feedback-driven process
• →The framework addresses technical challenges in grounding abstract feedback into spatially coherent planning actions
• →Graph-based experience banks enable systematic organization of dispersed residential feedback across urban contexts
• →Experimental results show consistent improvements over baseline methods on both conventional and experience-based metrics
• →The approach scales participatory planning engagement beyond traditional resource-intensive stakeholder consultation models