Real-Time Onboard Recognition of Gait Transitions for A Bionic Knee Exoskeleton in Transparent Mode

To achieve smooth locomotion transitions, locomotion intent prediction is very important for the control of knee exoskeleton. In this study, we develop a multi-sensor based locomotion intent prediction system based on Support Vector Machine (SVM), which can identify the current locomotion mode (sit, sit-to-stand, stand, level-ground walking, or stand-to-sit) and detect the locomotion transition between these modes onboard online. Two IMUs are mounted on the unilateral front of thigh part and shank part of the knee exoskeleton, and each of them generates 9 channels data. To evaluate the performance of this prediction system, several experiments are conducted on five healthy subjects. Average recognition accuracy is 96.89% ± 0.23%. Most transitions can be detected before the onsets of the transitions and no missed detections are observed for all the trials of the five able-bodied subjects.