Assistive Exoskeleton based on sEMG-based Motion Intention Recognition Study

To address the issue of imprecise recognition of human motion in upper extremity assistive exoskeletons, a method for predicting the motion intention of the human upper extremity using surface electromyography (sEMG) signals is proposed, tailored for a 6-degree-of-freedom (6-DOF) upper extremity exoskeleton system. First, a three-dimensional model of the exoskeleton was built through simulation experiments, and the exoskeleton's motion equation was established. The inverse kinematics (ikine) function was utilized to compute joint variables, while the ctraj and jtraj functions were applied to generate a smooth joint trajectory from the initial position to the target position, determining the motion process of the exoskeleton through trajectory planning. In terms of signal processing, a bandpass filter was constructed to filter the sEMG signals, removing noise and irrelevant information to obtain more accurate muscle signal features. Based on this filtered data, a convolutional neural network (CNN) model was constructed to predict and recognize the filtered sEMG signals. By leveraging the CNN's multilayer convolutional architecture, deep features were extracted from the signals, enhancing the model's ability to recognize human motion intentions. The simulation results demonstrated that the CNN model achieved an average recognition accuracy of over 90% for predicting human upper limb motion intentions, validating the feasibility and effectiveness of the proposed approach. This method not only improves the accuracy of motion intention recognition for exoskeleton systems but also provides significant technical support for future applications in upper limb rehabilitation and human-machine collaboration.