Enhanced Human-Robot Interaction for Robotic Container Unloading Systems Utilizing Digital Cousin and LLM

Existing robotic container unloading systems lack the adaptability and semantic reasoning required for robust performance due to dynamic environments and unstructured cargo arrangements. This paper proposes a novel LLM and Digital Cousins-driven robotic automation system to address these limitations. By leveraging the DeepSeek-V3LLM (large language model), the system integrates contextual understanding, real-time decision-making, and adaptive grasping strategies to handle irregular or occluded objects. A hybrid architecture synergizes LLM-based high-level planning with low-level robotic control, while a synchronized digital cousin framework ensures real-time physical-virtual synchronization. The digital cousin technique constructs semantically consistent virtual scenes from single RGB images using a 3D asset library, enabling cost-effective, automated synthesis while preserving core geometric and functional attributes. Experimental results demonstrate significant improvements in operational efficiency and robustness compared to conventional methods. This work bridges AI-driven cognitive capabilities with industrial robotics, offering a scalable and intelligent solution for autonomous logistics.