A Semi-active Back Assistive Exoskeleton With a Real-time Motion Classification Algorithm

Abstract

This paper presents a semi-active exoskeleton designed to support the user's back during lifting motions. A semi-active actuation system with a ratchet-pawl and a single-wire actuation mechanism was developed to provide significant assistive force with minimal energy consumption. The exoskeleton can generate an assistive torque of approximately 100 N <middle dot> m during lifting motions, while allowing natural motion during walking, standing, and sitting. A 1D convolutional neural network (CNN)-based real-time motion classification algorithm was developed to identify the user's motion in the early stages of the motion, which is an essential requirement for semi-active exoskeletons. To enhance the accuracy of the motion classification algorithm, the user's torso acceleration was used in addition to the hip joint angle. The proposed motion classification algorithm achieved an average accuracy over 95% in classifying lifting, sitting, standing, and walking motions, outperforming previous studies. The time required for classification was approximately 0.15 s, which is sufficient to control the semi-active mechanism by activating the supporting force in the early stage of motion.