Reciprocal human–machine learning flow modelling for assisted assembly systems

Industry 5.0 emphasizes human-centric manufacturing by integrating advanced assistive technologies for inclusive production environments. Large Language Models (LLMs) offer new possibilities for assembly error detection and correction. This study introduces a reciprocal human–machine learning framework utilizing LLMs vision capabilities to improve assembly accuracy through real-time image comparison and corrective instruction generation. Human and machine-in-the-loop mechanisms enable continuous refinement, minimizing labeled datasets while enhancing responsiveness. Experimental validation demonstrates the system's ability to detect errors, diagnose causes, and provide corrective actions. Findings show LLM-driven reciprocal learning improves efficiency, supports diverse operator needs, and enables adaptive error detection in manufacturing.