Unilateral Inertial and Muscle Activity Sensor Fusion for Gait Cycle Progress Estimation

This paper introduces a method which uses feedforward neural networks (FNNs) for estimating gait cycle progress using data recorded from inertial and muscle activity sensors attached to one side of the lower body. Three-axis inertial measurement unit (IMU) readings from accelerometers and gyroscopes located above the outer ankle and knee were fused with mechanomyogram (MMG) sensor readings from across major muscle groups on the left leg. Validation was against ground truth gathered concurrently with VICON motion capture. The performance was characterised by rms error (Erms) and max error (Emax), averaged across four cross-validated trials, and enhanced by adjusting number of sliding window frames and hidden layer neurons. The final configuration estimated gait cycle progress with Erms of 1.6% and Emax of 6.8%. This demonstrates promise for such a method to be used for control of unilateral robotic prostheses and exoskeletons, providing state estimation of gait progress from low power sensors limited to one side of the lower body.