A domain-informed vision-language model for sustainable freight: zero-shot classification of drayage truck powertrain and cargo types

Abstract Heavy-duty trucks contribute disproportionately to greenhouse gas (GHG) and air pollutant emissions, despite accounting for only a small share of vehicle fleets. Drayage trucks, a category of heavy-duty vehicles that transport containers between ports, railyards, and warehouses, are a major source of emissions in freight-intensive corridors, particularly around ports. Due to their short-haul and predictable operation cycles, drayage trucks are considered strong candidates for zero-emission technologies. As California and other states establish ambitious targets for transitioning drayage fleets into zero-emission vehicles, it is essential to assess fleet composition and the pace of technology adoption. Accurate identification of truck powertrain types, such as diesel, electric, hydrogen, and compressed natural gas (CNG), and cargo configurations is critical for tracking policy implementation, planning charging and fueling infrastructure, and reducing emissions from inefficient operations such as empty hauls. However, conventional classification methods rely on large manually labeled datasets, and existing studies have not focused on identifying truck powertrain types. This study introduces ZeroDray, a novel zero-shot classification framework that enables a domain-informed vision-language model to identify drayage truck attributes without requiring labeled training data. ZeroDray employs expert-informed prompts that integrate domain knowledge, visual evidence, spatial reasoning, and chain-of-thought processing to generate interpretable and accurate predictions. The framework was evaluated on 443 distinct drayage truck images collected along a highway corridor serving the Ports of Los Angeles and Long Beach and achieved F_1 scores above 92 percent across all 11 powertrain-cargo configurations. ZeroDray offers an interpretable framework to track zero-emission vehicle adoption, which supports data-driven decision-making in sustainable freight planning and environmental regulation.