ARTFEED — Contemporary Art Intelligence

LLM Multi-Agent Framework for Living-in-the-loop Urban Planning

ai-technology · 2026-05-27

LiPUP-MA, a novel multi-agent system, employs LLM-based agents to replicate residential environments and create feedback loops for participatory urban planning. This method tackles the cyclical aspects of real-world planning by shifting between simulated living scenarios and experience-informed plan adjustments. Major obstacles involve anchoring diverse living experiences within specific urban settings and converting subjective input into spatially coherent actions. The framework develops a Plan-centric Graph-based Experience Bank to systematically arrange urban-related feedback and incorporates spatially-constrained skill enhancement.

Key facts

  • LiPUP-MA is an LLM-based multi-agent framework for participatory urban planning.
  • It uses a closed-loop paradigm alternating simulated residential living and experience-driven plan revision.
  • Key challenges include grounding living experiences in urban contexts and translating feedback into spatially coherent actions.
  • The framework includes a Plan-centric Graph-based Experience Bank.
  • It uses spatially-constrained skill augmentation.
  • The approach addresses the cyclical nature of real-world planning.
  • Existing methods rely on static preference elicitation and one-shot discussions.
  • The paper is available on arXiv with ID 2412.20505.

Entities

Institutions

  • arXiv

Sources