ABSTRACT Urban logistics faces mounting pressure from growing demand and increasingly diversified service needs. However, traditional ground‐based delivery methods are becoming unsustainable, as they contribute to traffic congestion and environmental degradation. Urban rail transit, with its high punctuality, extensive coverage and lower passenger volumes during off‐peak hours, presents untapped potential for supporting logistics services in metropolitan areas. This study presents a metro‐based cross‐station delivery system built around autonomous delivery vehicles (ADVs). The proposed system features robust environmental perception through multi‐sensor fusion and YOLOv11s‐based object detection, with specific adaptations for detecting glass surfaces and handling boundary inflation constraints. It integrates Voronoi‐guided A * path planning with a proportional‐integral‐derivative (PID)–model predictive control (MPC) hybrid control framework to navigate the structural and operational complexities of metro stations. The system was validated through both simulation and real‐world deployment at five stations across Shenzhen Metro Lines 14 and 16. The ADV completed key tasks, including in‐station navigation, autonomous elevator interaction and train boarding and disembarkation. It also demonstrated reliable obstacle avoidance and consistent trajectory tracking under high passenger density, successfully fulfilling end‐to‐end delivery tasks. The results highlight the technical feasibility of metro‐integrated logistics and offer a practical reference for future research and deployment of urban delivery systems leveraging rail infrastructure.
Cao et al. (2026) studied this question.