A Smart Garment Workstation for Customized Apparel Production

In the garment production process, pattern makers are responsible for cutting and creating sample garments. Manual cutting is inefficient, and traditional cutting machines lack flexibility. In contrast, robotic arms are more efficient, flexible, and safer, while providing superior cutting quality. Based on these advantages, we developed a smart garment workstation for customized apparel production. This smart workstation is equipped with vision sensors, a smart screen, and robotic arms. To enhance automation in the robotic arm cutting process, we utilized a large language model (LLM) for code generation. Compared to machine learning classification methods and template-based generation approaches, the LLM offers exceptional generalization capabilities. It excels at interpreting perceived pattern information and accurately matching it with codes in the library, resulting in the generation of executable code for precise pattern cutting. Moreover, recognizing human behavior during the cutting process is crucial for safety. The unpredictable nature of human actions and the complexity of muscle parameters present significant challenges. To address this issue, we developed a human digital twin (HDT). This HDT model reduces uncertainty through diffusion and denoising processes and reconstructs accurate human geometry by applying diffusion constraints to the SMPL model. The established HDT acquires spatiotemporal data from vision sensors, recognizes human behavior. If the HDT model detects that the collision distance is below the threshold, it controls the robotic arm to prevent collisions and prompts the smart screen to provide collision warnings, ensuring the safety of the cutting process.