A machine learning based method of constructing virtual inertial measurement predictor of human body

In recent years, using human characteristics to assist exoskeleton robot control is one of the hot spots in the field of robot technology. For the problem that extremity installation of inertial measurement components in human body cannot achieve effective measurement, a method of constructing virtual inertial measurement predictor of human body based on machine learning is studied. The method uses the outputs of the inertial measurement components synchronously installed on the extremities and other parts of the body as the data samples, through recurrent neural network, it realizes the construction of virtual inertial measurement components and their predictors. In order to improve the training effect, the training samples are filtered based on gait phase detection. The simulation based on Anybody and MATLAB shows that, by installing the inertial measurement component near hip joint, the proposed method can effectively simulate and predict the inertial measurement component's output of centroid position of foot and surface of shank.