CityReal Framework for Human-Aligned Urban Behavior Simulation

CityReal improves urban simulation realism by modeling agents as intention-driven decision makers rather than step-by-step prompt followers. It uses learned textual adapters to align agent behavior modules with observed population statistics, preventing the reproduction of LLM behavioral priors.