A Study on the Exoskeleton Motion Intent Recognition Algorithm for Embedded Calculation

The exoskeleton robot to assist load-carrying has received much attention in recent years. It is a highly coupled human-machine system. In order to realize the compliant motion control target and complete the reliable power-assisted control for its wearer, it is necessary to accurately identify and predict the wearer's motion intention in real time. In this study, based on the foot pressure signal and the detected human motion information, the multi-sensor fusion method is used to complete the recognition of the wearer's motion intention. For the recognition of motion patterns, by comparing the recognition accuracy, resource consumption and real-time processing ability of various machine learning algorithms, the paper is finally determined that the support vector machine (SVM) is used to realize the action recognition for 8 daily motion patterns (Sit, Stand, Walk, Run, Ramp Ascent, Ramp Descent, Stairs Ascent and Stairs Descent), and the average recognition accuracy rate reaches 95%. For the prediction of motion phase and motion switching events, the neural-fuzzy inference method is used to complete the motion phase recognition and state switching event prediction. On the given test set, the accuracy of phase recognition is 99%, and the average absolute value of the deviation between the predicted state switching moment and the real value is around 61.6ms, which meets the requirements of exoskeleton compliance control for prediction time.