Database-driven approach for Biosignal-based robot control with collaborative filtering

In this study, we propose a databasedriven torque estimation approach for EMG-based robot control. For conventional EMG-based controllers, torque estimation models need to be carefully calibrated to control robots that have multiple degrees of freedom. However, such a calibration procedure requires significant effort and restricts the applications of EMG-based methods to practical situations. To cope with this issue, we use large-scale data acquired from other users to avoid the calibration process and propose collaborative filtering to estimate the joint torque of a new user by exploiting the previously derived relationships between the EMG signals and the joint torque of other users. To validate our proposed method, we compared the joint torque estimation performance with a standard linear conversion model. In our experiments, we controlled an upper-limb exoskeleton robot with the estimated joint torque where we used 16-ch electrodes to measure the EMG signals of subjects. In a comparison, our proposed method showed comparable control performance with the standard approach that requires a careful calibration process.