Real-Time Locomotion Mode Recognition and Assistive Torque Control for Unilateral Knee Exoskeleton on Different Terrains

Abstract

Stroke may lead to considerable physical impairment and functional disability, which affects walking ability. As a potential way to assist gaits, lower limb exoskeletons have been developed. To supply appropriate assistive torque, real-time accurate recognition of current gait mode is important. In this article, a hierarchical control system was proposed to recognize locomotion modes in real time for a unilateral knee exoskeleton on different terrains with specific assistive control strategies. Support vector machine classifier was used to recognize the locomotion mode based on two inertia measurement units. The corresponding assistive control strategy was designed according to the recognition result. Real-time recognition experiments under assistive torque control were conducted on five able-bodied subjects and one stroke patient, respectively. For the able-bodied subjects: first, no significance was found on the total recognition accuracies whichever the leading leg was for the five subjects (p = 0.057), which indicated the proposed method in this article was suitable whichever the leading leg was as far as the overall classification accuracy was concerned. Second, transitions occurred in swing phase when the leading leg was the paretic leg and transitions occurred in stance phase when the leading leg was the sound leg. No significance was found on mean delay time for the five subjects (p = 0.785) whichever the leading leg was, which indicated that the proposed method in this article was suitable for these two leading legs as far as the mean delay time were concerned. Third, the method of generating the assistance based on the previous gait cycle time was demonstrated to be reasonable and the tracking performance of the torque could meet the requirement. For the stroke patient, similar experimental results were obtained.