Motion-based Grasp Selection: Improving Traditional Control Strategies of Myoelectric Hand Prosthesis

This paper introduces a novel prosthetic hand control architecture using inertial information for grasp prediction in order to reduce the cognitive burden of amputees. A pair of inertial measurement sensors (IMUs) are fitted on the wrist and bicep to record arm trajectory when reaching to grasp an object. Each object class can be associated with different methods for grasping and manipulation. An observation experiment was conducted to find the most common grasping methods for generic object classes: Very Small (VS), Small (S), and Medium (M). A Cup (CP) class was also examined to find differences in grasping habits for pick and place, and drinking applications. The resulting grasps were used to test the discriminatory ability of inertial motion features in the upper limb for VS, S and CP object classes. Subject experiments demonstrated an average classification success rate of 90.8%, 69.2% and 88.1% for VS, S and CP classes respectively when using a k-nearest neighbors algorithm with a Euclidean distance metric. The results suggest that inertial motion features have great potential to predict the grasp pattern during reach, and to the authors' knowledge, is the first IMU-based control strategy to utilize natural motion that is aimed at hand prosthesis control.