Predicting elbow motion intention based on different electromyography and inertial measurement unit sensor fusion structure

As information on motion intention prediction methods, bioelectrical and physical signals have been commonly used. However, both types of signals have opposing weaknesses and strengths. To compensate for these limitations, many studies have fused and utilized both signal types, but they have rarely discussed how to fuse them in terms of input/output structure, despite the significant impact of such discussion on prediction performance. Therefore, in this study, we designed and analyzed various sensor fusion structures using electromyography (EMG), one of the bioelectrical signals, and inertial measurement unit (IMU) signal, one of the physical signals, and then determined an optimal structure for using in our prediction model. To predict future motion intention in advance, the concept of the response time difference between EMG and IMU signals was employed in artificial neural network (ANN) training. Various experiments with a simple motion and two various motion scenarios were conducted with three subjects to verify the effectiveness and robustness of the proposed method. The results show that proposed method can predict future elbow angles with high accuracy and performance consistency across all subjects. Furthermore, these results allow joint angle synchronization of robot and human, and consequently reduce the discomfort of the subject from a muscle usage perspective.