Wrist angle estimation based on bilinear model using electromyographic signals from circularly-attached electrodes

This paper aims at estimating a joint angle from EMG signals. A bilinear model reflecting a muscle viscous-elastic property changing with its activity is introduced as the dynamic equation. Learning parameters of the dynamical model will decrease the effect from the difference of human users. The signal separation techniques based on the principal or independent component analysis are adopted to solve a problem about noises and crosstalk in EMG measurements, by applying them to some signals detected from some numbers of electrodes attached around the body parts. Experiments on the hand dorsal extension/palmar flexion evaluates some estimation results of angle deviation in the actual hand movements.