CityReal Uses LLM Agents to Improve Urban Simulation Alignment with Human Behavior

CityReal, a modular framework introduced in an arXiv preprint (2608.16897), addresses limitations in large-scale urban simulation using large language model (LLM) agents. Existing methods rely on few-shot prompting, causing agents to reproduce LLM priors rather than target population behavior. CityReal models agents as intention-driven decision makers pursuing coherent mobility and activity plans. These agents adapt over time by learning habits from experience and constraints. To improve population-level realism, the framework learns textual adapters aligning agent decisions with observed population statistics. The experiments demonstrate improved alignment with real-world human behavior at both micro and macro levels. The paper is available at arxiv.org/abs/2608.16897.