자율 로봇 제조를 위한 계층적 작업 인식 프레임워크

Abstract

This study proposes a hierarchical task recognition framework that enables robots to precisely perceive and interpret their actions in complex and unstructured manufacturing environments. The system decomposes entire processes into modular primitive tasks and primitive actions, leveraging multimodal sensor data to evaluate task states in real time. By structuring each task stage as an independent, reusable module, the framework supports flexible composition and adaptation across diverse process scenarios. It further incorporates both success and failure conditions to ensure robust control under unexpected circumstances. Crucially, the system allows robots to dynamically adjust their execution based on real-time feedback, enhancing autonomy and responsiveness. The framework has been validated in real-world assembly tasks, demonstrating its ability to adapt to various task types and conditions without structural modification. This approach offers a practical foundation for intelligent robotic systems capable of real-time understanding and adaptive control in dynamic manufacturing settings.