A Machine Learning Approach to Hand-Arm Motion Prediction for Active Upper Extremity Occupational Exoskeleton Devices

Abstract

Exoskeleton devices are currently being utilized in a variety of occupational settings to reduce musculoskeletal efforts to lower fatigue, improve performance, and minimize work-related injuries associated with musculoskeletal disorders (MSDs). The intrinsic challenges that accompany the development of fully supporting and active upper extremity multi-degrees of freedom (DOF) devices include not only the mechanical design, but also lack of an intuitive way to control and operate these devices. A proof-of-concept instrumented handle consisting an embedded sensor network was developed with the intention to utilize artificial neural networks (ANN) to properly identify the intended motion of the user and to estimate the motion intensity. Results show this method is feasible and delivers useful insight into developing the next stages of the “smart handle” technology that will include the remaining hand motions, correctly classifying combination of intended motions and using the handle output to control complex multi-DOF upper extremity exoskeletons devices.