A compact and cost-effective pattern recognition based myoelectric control system for robotic prosthetic hands

In this study, we propose, implement and test a compact pattern recognition (PR) based myoelectric control system to operate robotic prosthetic hands. Instead of using a bulky sensor configuration (8 or more sensors on the forearm) as most commercial or academic PR systems use, only two sEMG-IMU sensors are deployed on the forearm, consistent with the existing myoelectric hands' electrode configuration, to bring the ease of implementation and consequently improve its practical application. For maintaining the recognition performance as much as possible, two additional sEMG-IMU sensors are attached to the upper arm, which aims to collect more information of muscular activities for the benefits of classification but not affecting the existing forearm sockets of the prosthetic hand users. The offline analysis by using the data from 10 intact subjects and 10 transradial (i.e. below-elbow) amputees shows the comparable recognition performance using the proposed sensor configuration (2 + 2 sensors) and classification technique (Random Forest classifier) to that using the full pack of sensing system (10 sensors on the forearm + 2 sensors on the upper arm). In the online tests, the proposed PR system was employed to control the UoW/ACES soft robotic prosthetic hand to successfully and reliably perform seven common wrist/thumb movements for the activities of daily life (ADL).