Prediction of Joint Angles for Human Elbow Motion Based on sEMG

In the field of intelligent rehabilitation, the primary reliance is on rehabilitation robots, which often struggle to continuously respond to a patient's movement intentions in real-time and with accuracy. To improve the level of human-robot interaction (HRI) and provide personalized rehabilitation exercises with active patient participation, this study developed a joint angle prediction model using surface electromyography (sEMG). Initially, six healthy subjects were selected, and devices for sEMG and angle acquisition were employed to synchronize the collection of sEMG and elbow joint angle data for each participant. Subsequently, the preprocessed data were input into the designed Transformer-based joint angle prediction model. Finally, the performance of the Transformer-based model was compared with that of Recurrent Neural Networks (RNN) and Convolutional Neural Network-Long Short-Term Memory (CNN-LSTM) models. Experimental results indicate that the Transformer-based joint angle prediction model outperforms its counterparts, achieving a Mean Absolute Error (MAE) of less than 0.065, thereby suggesting its potential for future applications.