Beyond the Legs: Leveraging Trunk and Driving Muscle EMG to Improve Gait Model Predictions.

Gait prediction models based on electromyography (EMG) have been extensively used for motion intent detection in lower-limb exoskeletons. Traditionally, gait prediction models tend to rely solely on lower-limb EMG signals. However, based on the importance of trunk muscles in clinical gait studies, we hypothesise that these could provide valuable information for the development of more accurate gait models. Furthermore, based on previous literature on muscle functionality during steady-state walking, we also hypothesise that focusing on muscle groups that are responsible for movement generation in gait (referred to as driving muscles, or "drivers") will lead to improved kinematics predictions compared to other muscle subsets. In this study, we examine the predictive value of EMG signals extracted from the trunk and driving muscles versus purely the lower-limbs in predictive gait models. We investigate prediction horizons of 0.05 ms (one time-step ahead), 100 ms, and 200 ms, and predict the angular output of the hip, knee and ankle joints, as well as the estimated gait cycle percentage. Our results show that models trained using trunk muscle EMG significantly outperform models trained only on lower-limb EMG, particularly for knee angle outputs with longer prediction horizons $(p<0.05)$. Furthermore, training on selected driving muscles further improved model performance when predicting the knee and ankle joint kinematics, particularly with short prediction horizons ($p<\mathbf{0. 0 0 1}$). These findings support the hypothesis that including trunk muscles contributes significantly to the development of accurate gait models, and highlight the potential of targeting "drivers" to select more relevant EMG inputs for kinematic predictions.