A Method of Motion Information Overall Sensing of Wearable Robot Based on Virtual Inertial Measurement Unit

A machine learning based method is studied for motion information comprehensive sensing of medical rehabilitation wearable robot. Taking use of human kinematics model to detect sample data, the model combines a type of visual-geometry-group-network and a long-short-term-memory neural network model (VGG-LSTM) to generate virtual inertial measurement unit. It is meant to realize the method of motion information overall sensing with a relatively small number of inertial measurement units. Furthermore, how to balance the complexity of machine learning model and the approximation accuracy of virtual inertial units becomes the focus. By analyzing the effects of VGG and layers of LSTM on the approximation accuracy of virtual inertial information and the utilization rate of computer system resources, the optimal synthesis of the neural network model structure. It both satisfies the comprehensive index of performance and system complexity optimization. The neural network structure lays a foundation for the engineering application of motion information comprehensive sensing of the wearable robot.