Finger Position and Force Simultaneous Prediction Using A-mode Ultrasound

This study attempts to estimate hand finger movement and force through forearm muscle deformation for muscle-driven dexterous human-machine collaboration scenarios based on A-mode ultrasound sensing approaches. Six healthy subjects were recruited to participate in the middle finger movement experiment. The least squares support vector machine (LS-SVM) regression model and Gaussian mixture regression (GMR) model were developed to estimate hand finger movement and force for datasets captured from the flexor digitorum superficialis (FDS) muscle belly, extensor digitorum (ED) muscle belly. The average results revealed that GMR outperformed LS-SVM in simultaneous estimation for finger movement $(R^{2}=0.919, N RMSE = 9.89{\%}$ for GMR and $R^{2} = 0.909, NRMSE=10.41{\%}$ for LS-SVM) and finger force $(R^{2}=0.889, NRMSE = 10.34{\%}$ for GMR and $R^{2}=0.865, NRMSE = 11.40{\%}$ for LS-SVM). Besides, the GMR performance on finger movement estimation is better than that of finger force estimation $(p . These results demonstrated the feasibility of finger force and joint angle simultaneous estimation based on A-mode ultrasound, demonstrating the potential for tactile driven human-machine collaborative systems applications.