Bridge: AI Framework for Urban Delivery Demand Forecasting

A new framework named Bridge has been developed by researchers to predict urban delivery demand, especially in cold-start areas where historical data is absent. This retrieval-augmented spatiotemporal graph framework integrates an inductive contextual graph backbone with a memory component that is sensitive to time across region-time windows. Bridge leverages regional context and recent trends to extract future demand patterns from memory for each target area, enhancing the backbone forecast via a gated fusion mechanism. This innovative method overcomes the challenges faced by current spatiotemporal forecasters, which often fail to capture short-term operational dynamics in unfamiliar service zones. The findings are published in arXiv preprint 2605.19172.