Multi-class SVM Based Real-Time Recognition of Sit-to-Stand and Stand-to-Sit Transitions for a Bionic Knee Exoskeleton in Transparent Mode

• Published in: Intelligent Robotics and Applications Vol. 10462; pp. 262 - 272
• Main Authors: Liu, Xiuhua, Zhou, Zhihao, Mai, Jingeng, Wang, Qining
• Format: Book Chapter
• Language: English
• Published: Switzerland Springer International Publishing AG 2017; Springer International Publishing
• Series: Lecture Notes in Computer Science
• Subjects: Gait transitions; Knee exoskeleton; Locomotion intent prediction; Multi-class SVM
• ISBN: 9783319652887; 3319652885
• ISSN: 0302-9743; 1611-3349
• DOI: 10.1007/978-3-319-65289-4_25

Real-time locomotion intent recognition is a challenge in lower-limb exoskeletons. In this paper, we present a multi-sensor based locomotion intent prediction system for sit-to-stand and stand-to-sit transition recognition the subject wears a knee exoskeleton in transparent mode. The desired reference torque for movement control is obtained from a direct torque control loop. The feedback torque is estimated by an inner current control loop. Five able-bodied subjects were recruited in experiments. The classifier is based on multi-class Support Vector Machine. Four kinds of modes and four kinds of transitions are tested in this study. Recognition accuracy during steady periods is 99.68% ± 0.07% for five able-bodied subjects. And during transition periods, all the transitions are correctly detected and no missed detections was observed for all the trials of the five subjects. Preliminary experimental results show that the proposed method is capable of performing real-time intent recognition and consequently reduces the interaction force between human body and the exoskeleton.