A Gait Generator for Guiding Knee Joint Motion During Overground with a Exoskeleton Robot

Abstract

Under the background of the rapid development of artificial intelligence technology, the lower limb exoskeleton robot technology has been further developed. However, due to the complexity of the perception system and the uniqueness of human gait patterns, it is difficult for the exoskeleton to achieve good coordination with the wearer, which is challenging for the wide application of exoskeleton robots. We develop a neural network model based on inertial sensor data to estimate gait trajectories in real time. First, using the Butterworth filter to remove high-frequency noise from the inertial signal. Then, the designed network model is applied to establish the nonlinear mapping between the motion features and the motion trajectory of the knee joint. Experimental results show that it can adapt to dynamic speeds from 0.8 to 2.4 km/h. The model’s predicted gait patterns closely matched the subjects’ actual joint trajectories, with a Pearson correlation coefficient as high as 0.963. Compared with the LSTM method, the threshold absolute deviation and mean absolute deviation are reduced by 73.41% and 63.45%, respectively. Our study validates the feasibility of using a deep learning model based on IMU sensor data to accurately predict the wearer’s gait trajectory and improve the wearable comfort of the powered lower limb exoskeleton robot.