Surface Electromyography-Based Wrist Angle Estimation and Robotic Arm Control with Echo State Networks

Continuous estimation of joint angles based on surface electromyography (sEMG) signals is a promising method for naturally controlling prosthetic limbs and assistive devices. However, conventional methods based on neural networks have limitations such as long training times and calibration burdens. This study investigates the use of an echo state network (ESN), which enables fast training, to estimate wrist joint angles from sEMG. Five participants mimicked the motion of a 1-degree-of-freedom robotic arm by flexing and extending their wrist, while sEMG signals from the wrist flexor and extensor muscles and the robotic arm’s angle were recorded. The ESN was trained to take two sEMG channels as input and the robotic joint angle as output. High-accuracy estimation with a median coefficient of determination R2 = 0.835 was achieved for representative ESN parameters. Additionally, the effects of the reservoir size, spectral radius, and time constant on estimation accuracy were evaluated using data from a single participant. Furthermore, online estimation of joint angles based on sEMG signals enabled successful control of the robotic arm. These results suggest that sEMG-based ESN estimation offers fast, accurate joint control and could be useful for prosthetics and fundamental studies on body perception.