A structured hierarchical model for human-robot collaboration in industrial assembly

Global manufacturing is shifting toward customized, small-batch products, which demand production systems that are both flexible and agile. Assembly is a critical stage and must remain both adaptable and efficient. Human–robot collaboration (HRC) offers a path forward by combining human adaptability and dexterity with robotic efficiency and repeatability, but practical deployment still struggles to ensure safety and efficiency. These challenges arise from the complexity and dynamics of tasks, human behavior, and robot behavior. Prior research has advanced task planning, but has only weakly integrated how humans execute tasks and how robots should respond accordingly. Studies of human behavior often isolate specific aspects (e.g., action recognition, motion capture, or prediction), and corresponding robot control strategies are typically case-specific. As a result, the interplay among task structure, human behavior, and robot behavior remains insufficiently organized.
A structured model for HRC in industrial assembly is needed to clarify task assignment and reinforce task scheduling, support a structured understanding of human behavior, and assist dynamic robot control strategies. Therefore, this thesis aims to build a structured hierarchical model for HRC to address both safety and efficiency in industrial assembly. The study adopts the Design Science Research Methodology (DSRM). The designed model integrates a task hierarchy combined with the corresponding human- and robot-behavior hierarchies. The development process incorporates: human task monitoring based on object tracking and Human–Object Interaction (HOI) analysis; atomic action identification through segmentation and clustering methods; Human Motion Prediction (HMP) using a Transformer-based model.
The effectiveness of the hierarchical HRC model in controlling robot systems across the subtask, atomic action, and motion frame levels has been demonstrated through three application scenarios: dynamic task scheduling driven by real-time subtask status of the human; adaptive robot control enabled by identified atomic action progress; and proactive robot control based on predicted human motion. The results prove that the proposed hierarchical HRC model enhances both human safety and process efficiency in industrial assembly tasks.
The structured hierarchical HRC model manages task complexity, supports multilevel human behavior understanding, and enables organized robot control. The approach aligns with the Industry 5.0 paradigm by promoting human-centric production and supports sustainability through improved economic efficiency and social inclusion. Future work will extend the human model to perception and cognition, incorporate user studies to assess subjective experience, and explore richer communication and integration with emerging technologies.