UrbanAgent Uses Multi-Agent AI for Enhanced Urban Profiling.

Urban region profiling, which is crucial for applications like population estimation and economic assessment, traditionally relies on fusing diverse urban data into latent embeddings. However, these correlation-driven methods often struggle with cross-modal inconsistencies and lack robustness in varied or unseen urban environments. A novel agentic framework called UrbanAgent redefines this task as a reasoning-driven inference problem. It assigns an independent AI agent to each data modality (e.g., satellite imagery, points of interest, textual descriptions) and facilitates structured collaboration among them to explicitly resolve inconsistencies. UrbanAgent also incorporates active evidence acquisition and iterative reasoning, allowing agents to verify uncertain inferences by retrieving external knowledge through tools, optimized via reinforcement learning. Extensive evaluations on global urban datasets demonstrate that UrbanAgent consistently surpasses existing baselines in accuracy and shows strong generalization capabilities in previously unobserved cities.