Arm Movements Effect on Grasping Force Prediction Using Surface Electromyography Signals

This study investigated the arm movements effect on the relationship between surface electromyography (EMG) signals and grasping force. An experiment was conducted with four static arm conditions and two dynamic arm conditions. Six able-bodied subjects participated in the experiment. Surface EMG signals were acquired from five forearm muscles to build a multiple linear regression model. Subjects were instructed to complete three kinds of calibration tasks to train the model and one voluntarily varying grasping force task to test the model performance. The grasping force exerted by each subject was limited to be lower than 50% maximum voluntary contraction (MVC) grasping force. Mean absolute difference (MAD) between predicted and observed grasping force was used to estimate the prediction performance. The window size of moving average filter was firstly optimized. Results showed that arm movements had a significant impact on grasping force prediction performance. Inter-condition MADs (training data and testing data are from different arm conditions) were greater than intra-condition MADs (training data and testing data are from the same arm condition, average 7.41% +/- 1.46% MVC vs. 6.03 +/- % 0.40% MVC, p = 0.023). A multi-condition training scheme was applied to attenuate the arm movements effect. The multi-condition training scheme was proved to be useful to improve the model robustness to the arm movements effect.