Simultaneous Estimation of Hand Joints’ Angles Toward sEMG-Driven Human–Robot Interaction

Human beings have very dexterous hands which help us manipulate a lot of complicated tools. But most researches focus on gesture recognition. This situation leads to the phenomenon that our prosthetic hands lack continuous and natural movement. This paper proposes a method aimed to decode finger movements continuously and simultaneously based on multi-channel surface electromyography signals. This algorithm is useful for surface electromyography(sEMG) decoding controlled for a prosthetic hand, exoskeleton robots for upper-limb rehabilitation, and remote mechanical hand control. Firstly, the feature extractor based on a sliding time window is used to extract 8 kinds of sEMG features (mean absolute value, integral sEMG value, root mean square, waveform length, logarithmic feature, zero-crossing points, and slope symbol change) from the sEMG signals of 8 channels of the forearm. The estimated angle of the metacarpophalangeal joint with large fluctuation is optimized by inputting them to the deep forest regression model; Then, the artificial neural network is used to optimize this estimated angle, to create a comprehensive regression model combining the deep forest regression model and artificial neural network; Finally, the comprehensive regression model is used to continuously and accurately decode the collected surface EMG signal to obtain the prosthetic hand finger joint angle control values, and the other finger joint angles can be obtained through proportional control principle. The experimental results show that the average trajectory tracking accuracy of the proposed method is 42% higher than that of the traditional Gaussian process method, reaching 84.4%, which proves that the proposed method has a very good effect on finger joint angle estimation based on sEMG signal.