Shared Control of Upper Limb Prosthesis for Improved Robustness and Usability

Feed-forward control paradigms currently dominate upper limb prosthesis research. Electromyographic (EMG) signals measured from muscles in a user's residual limb are filtered and processed before becoming the input to a machine learning algorithm. The output of this algorithm is sent directly as a command to a prosthetic. Despite advances in feed-forward methods, upper limb prostheses remain difficult for amputee users to control, and abandonment rates are high [1]. In this paper, we present a novel shared control paradigm that uses a hybrid gaze/EMG interface to control prosthetic hand and wrist movements. Six subjects used a virtual prosthesis to perform a pick-and-place task in an augmented reality (AR) environment using both a semi-autonomous (SA) controller and a feedforward (FF) controller representative of the current state of the art. Results show a 75% increase in average successful task completion rate when using the SA controller instead of the FF controller. The SA controller was found to be more robust against variation in object type and orientation, scored significantly higher on subjective usability metrics (p <= 0.05), and resulted in a dramatic decrease in user frustration during the task (p <= .01).