Deep Reinforcement Learning–Empowered Digital Twins for Carbon‑Efficient and Data‑Driven Logistics Decisions

Reducing carbon emissions in logistics operations has become a critical challenge for organizations seeking sustainable and data‑driven supply chain management. Traditional routing and operational planning methods often fail to adapt to dynamic logistics environments and real‑time environmental considerations. This study proposes an intelligent decision‑making approach that integrates digital twin technology with deep reinforcement learning to support carbon‑efficient logistics operations. The digital twin environment represents real‑time logistics activities and enables continuous interaction between physical delivery processes and data‑driven decision models. Within this environment, a Deep Q‑Network (DQN)–based learning mechanism is developed to optimize routing and operational choices while considering energy consumption and carbon emission impacts. A simulation experiment based on an urban last‑mile delivery network with 40 customers and 10 delivery vehicles is designed to evaluate the proposed approach under realistic operational conditions. The results demonstrate that the learning‑enabled digital twin can improve routing efficiency while simultaneously reducing carbon emissions compared with conventional decision strategies, achieving approximately 9.5% improvement in operational performance and 7.4% reduction in carbon emissions. The findings highlight the potential of integrating deep reinforcement learning and digital twin technologies to support carbon‑aware and adaptive logistics decision making in modern supply chains.