Exploring possibilities for real-time muscle dynamics state estimation from EMG signals

Real-time estimation of muscle state (activation, force and length) can be used for visualization of muscle function in neuromuscular rehabilitation and control of wearable assistive and augmenting devices. This paper proposes two formulations for developing linear state estimations of muscle dynamics in isometric contractions: the Luemberger observer and the Kalman filter. One of the main requirements of an estimator for such application is small lag, and a Finite Impulse Response (FIR) pre-filter was used to eliminate part of the high-frequency content of the EMG signal. A run test with human soleus EMG sample was performed. Both Luemberger observer and Kalman filter provided reasonably accurate estimations of the muscle state, compared to usual off-line EMG-driven analysis. In addition, the role of some Kalman filter parameters is analyzed.