Interlimb Gait Trajectory Synchronization from Inertial Measurement Unit (IMU) with ResUNet

Success in gait rehabilitation is one of the most important factors determining the quality of life in patients with neuronal injuries. Although all four limbs are involved in human gait physiology, most rehabilitation approaches separate the upper limbs from the gait rehabilitation system. We proposed autoencoder-based networks with residual blocks, ResNetAE and ResUNet, for an interlimb gait trajectory synchronization system for the purpose of active rehabilitation system control. Gait data was collected from healthy subjects (N=30), consisting of early adulthood (18-35 years, N=10), middle adulthood (36-60 years, N=10), and elderly individuals (over 61 years, N=10). Leveraging the Xsens MVN Inertial Measurement Unit (IMU) system, the positions of each body segment were extracted and synchronized in pairs. Utilizing the X and Z coordinate positions from the hands and feet, the model showed promising results in predicting two pairs of ipsilateral and contralateral interlimb trajectories, achieving the best Mean Absolute Error (MAE) of 0.0472 meters, a time error on the X-axis of 33.33 milliseconds, and a Phase Synchronization Index (PSI) of 96.70% when compared with the LSTMAE baseline (0.0547 meters, 33.33 milliseconds, and 92.83%) on unseen subjects. The models' performance is sufficient for further interlimb phase synchronization studies and for investigating their capability in the interlimb-aware gait rehabilitation exoskeleton systems.