A new Kinect-based guidance mode for upper limb robot-aided neurorehabilitation

During typical robot-assisted training sessions, patients are required to execute tasks with the assistance of a robot while receiving feedback on a 2D display. Three-dimensional tasks of this sort require the adoption of stereoscopy to achieve correct visuo-motor-proprioceptive alignment. Stereoscopy often causes side-effects as sickness and tiredness, and it may affect the processes of recovery and cortical reorganization of the patients' brain in an unclear way. It follows that it is preferrable for a robot-assisted neurorehabilitation therapy to work in a real 3D setup containing real objects rather than using virtual reality. In this paper, we propose a new system for robot-assisted neurorehabilitation scenarios which allows patients to execute therapy by manipulating real, generic 3D objects. The proposed system is based on a new algorithm for identification and tracking of generic objects which makes efficient use of a Microsoft Kinect sensor. We discuss the results of several experiments conducted in order to test robustness, accuracy and speed of the tracking algorithm and the feasibility of the integrated system.