A Driving Method for Medical Robot Based on CNN-Attention Continuous Myoelectric Prediction

This study presents an EMG-driven prediction model for optimizing actuation methods in medical robots, employing multi-scale convolution and multi-head attention mechanisms to capture temporal features in electromyographic (EMG) signals. Experiments on the Ninapro dataset compared the proposed CNN-Attention model with LSTM networks using surface EMG (sEMG) data and kinematic evaluations of six hand grasp movements. Results demonstrated that CNN-Attention outperformed LSTM in all performance metrics, achieving superior accuracy and computational efficiency for precise hand motion control in medical robotics applications.