A Framework for Human-Robot Interaction in Industrial Environments

Abstract

With the advantages of human-centric industrial environments, developing new technologies for human-robot interaction (HRI) has become an important research frontier for advancing the capabilities of industrial facilities. However, based on the literature, there is a lack of safe and reliable industrial collaborative robots capable of performing independent tasks and working collaboratively with a human. To address this gap, this thesis proposes a novel framework version for HRI suited for industrial environments. The framework integrates a form of perception, decision-making, and control to enable structured and, thereby, safe interaction between humans and robots. In addition, the design is suited for relatively cost-effective equipment commonly found in industrial settings.

First, a custom dataset is created to represent established categories of interaction modes, tasks, and actions relevant to industrial settings. Gesture recognition is achieved through human pose estimation and the formulation of distinguishing geometric relationships between skeletal keypoints, combined with image classification of a physical identifier to associate the robot with the intended human collaborator. Together, the perception components enable a lightweight and effective method of human-robot communication and task identification. Based on the dataset size, the evaluation of the gesture recognition strategy shows relatively good generalization to unseen participants. Then a gesture-based switching finite state machine (FSM) is designed to enable structured human–robot communication. Together with the perception components, clear decision-making can occur to enable a safe protocol-driven HRI.

Second, a path following control strategy suited for human-robot collaboration (HRC) is designed in the robot control. Implementing impedance control realizes a compliant robot behaviour for human-robot collaborative tasks, where the human, with expert contextual perception and cognitive on-demand adaptation, can guide robot motion. By mapping the robot coordinates into tangential, transversal, and orientation states relative to a nominal path, the approach potentially allows direction-specific motion objectives unique to the task.

Finally, the gesture recognition, FSM decision logic, and path following robot control are integrated into an actionable framework. The resulting system enables clear protocols to facilitate safe and effective HRI. Experimental validation demonstrates successful HRI, human-robot collaborative tasks, and highlights the potential tailoring of the control to different collaborative task conditions.