Background
Conventional robotic assembly relies heavily on engineers for programming, path setup, and repeated tuning, resulting in high deployment costs and limited flexibility when adapting robots to new production lines or tasks. This research uses human demonstrations as the source of task knowledge, aiming to transform unstructured human operations into structured and executable robot task workflows, thereby improving the efficiency of robot deployment and task adaptation in smart manufacturing scenarios.
Research Objectives
To develop a demonstration-driven intelligent assembly knowledge translation system that enables robots to understand assembly procedures from human demonstrations, parse atomic actions, operation sequences, and key parameters, and automatically generate executable robot assembly instructions, thereby lowering the barrier for robot deployment and production-line adaptation.
Methods
The proposed method integrates human operation demonstrations, visual perception, action understanding, atomic action libraries, and modular low-level skills to decompose continuous and unstructured demonstrations into executable assembly subtasks. The system further employs digital-twin and high-fidelity simulation environments to validate state transitions, positioning accuracy, task planning, and low-level execution results, providing the foundation for subsequent deployment on physical robots and real production-line validation.
Innovation
The innovation lies in moving beyond manually programmed robot workflows by automatically translating operational knowledge embedded in human demonstrations into structured procedures that robots can understand and execute. By combining visual perception, task-level semantic parsing, atomic action representation, and digital-twin validation, the proposed system improves the repeatability, scalability, and field adaptability of intelligent assembly processes.
Expected Outcomes
The expected outcomes include a demonstration-driven intelligent assembly knowledge translation system, including human demonstration parsing, atomic action recognition, task-parameter extraction, robot assembly instruction generation, and digital-twin validation workflows.