Real-time onboard SVM-based human locomotion recognition for a bionic knee exoskeleton on different terrains

Abstract

Locomotion intent recognition is essential for human-robot interaction to realize assistance control. In this study, we proposed a real-time locomotion intent recognition method running on the exoskeleton control hardware, which can recognize current locomotion mode and detect locomotion transitions in advance. Signals from two inertial measurement units installed on the exoskeleton (one on the thigh part, the other on the shank part) are used to detect the human intents, which include three locomotion modes (level walking (LW), stair ascending (SA) and stair descending (SD)) and four transitions (LW → SA, SA → LW, LW → SD and SD → LW) in this study. For a unilateral exoskeleton, the leg wearing the robot system played different roles especially for locomotion transitions and corresponding experiment trials are conducted respectively. Online recognition accuracy during steady locomotion periods is 95.74± 2.19%. During locomotion transition periods, all the transitions are correctly detected and most of them can be detected before transiting to new locomotion modes.