Real-Time Gait Intention Recognition for Active Control of Unilateral Knee Exoskeleton

Real-time gait estimation is important for the synchrony control of robotic exoskeleton to provide walking assistance. However, for stroke patients with hemiplegic paralysis, the gait pattern is very complex. Accurate and timely gait intention recognition is therefore difficult. To achieve human-robot synchrony control for an unilateral knee exoskeleton, a gait intention recognizer coupling the adaptive frequency oscillator (AFO) and back propagation neural networks (BPNN) is proposed in this paper. The BPNN is trained with gait data of healthy subjects and stroke patients to improve the accuracy of recognized gait pattern, which is then imported into flexible interaction module to provide appropriate assistance. To evaluate the performance of gait intention recognition, three stroke patients were recruited to conduct level ground walking tests. The kinematic and biomechanical data were captured in each test and processed for the evaluation. Experimental results demonstrate the effectiveness of gait intention recognition and movement assistance.