Straight Walking Gait Parameter Prediction with a Reduced Number of Sensors

The correct prediction of gait parameters is crucial in hip exoskeleton-assisted walking tasks, which can reduce the fall rate of individuals wearing these devices. Current prediction methods focus primarily on the foot and shank segments and do not take into account the lower back and thigh segments. This paper introduces a novel LSTM-based gait parameter prediction model for straight walking paths that uses data from IMU sensors mounted on the lower back and thigh to predict three key gait parameters: step distance, swing height, and swing velocity. Two approaches were tested: individual LSTM models for each gait parameter and a unified LSTM model that predicted all three gait parameters at once. Results in a 10% accuracy range demonstrate sufficient estimation accuracy, even when using a reduced number of sensors and highlight the superior performance of individual LSTM models over the unified LSTM model.