Deep Learning Technique to Identify Abrupt Movements in Human-Robot Collaboration

Abstract

According to the ninth Sustainable Development Goal (SDG9) of the 2030 Agenda, an upgrade of technological capabilities promotes the development of a sustainable and inclusive industrialization. In this context, a fundamental requirement is represented by the operator’s safety inside the workspace, especially when it is shared with a collaborative robot. Even if typical collaborative tasks are usually characterized by repetitive and controlled kinematics and dynamics, external disturbances and environmental factors can make the operator executing unexpected and abrupt gestures which are highly variable. The current study aimed at identifying human unexpected movements measuring upper body accelerations through wearable magneto-inertial measurement units. An experimental pick and place task was performed by five subjects, combining both routine and abrupt movements. Recurrent neural network was exploited to distinguish between normal and unexpected gestures. Overall, the chosen deep learning network and the developed pre-classification method for accelerations proved to be suitable for the identification of human abrupt movements in interaction with machines.