Automating the assessment of wrist motion in telerehabilitation with haptic devices

Abstract

Stroke-induced motor impairment often prevents survivors from participating in activities of daily living, adversely impacting their quality of life. Desktop delta robots such as the Novint Falcon have been utilized in various home-settings to help recover fine-motor skills. They are compact and affordable, and can provide programmable sensorimotor feedback. In spite of these favorable features, it is presently not possible to directly measure the user's wrist angles while interacting with these robots, which undermines their prospective use in telerehabilitation as patients' motor performance cannot be reliably assessed. Here, we propose an experimental set-up where patients strap a smartphone device to their forearm and manipulate a haptic robot. In this setting, data from inertial sensors embedded in the smartphone will be integrated with data from the robot in a classification algorithm that infers the wrist angle. To study the viability of this approach, we perform experiments with one healthy user. We fix two inertial measurement units on their body, one on their forearm and one on the back of their hand, to measure the true wrist angle as they perform a motor task with a Novint Falcon device. We train a machine learning algorithm that predicts wrist angles from a single wearable sensor and the Novint Falcon movements. This effort constitutes a step toward automatic assessment of wrist movements in fine motor telerehabilitation and could enable real-time feedback in the absence of a therapist.