Online adaptive assistance control in robot-based neurorehabilitation therapy.

Repetitive and task specific robot-based rehabilitation has been proved to be effective for motor recovery over time. During a therapy, the task should improve subject's impaired movements, but also enhance their efforts for a more effective recovery. This requires an accurate tuning of the task difficulty, which should be tailored directly to the patient. In this work, we propose a system for real-time assistance adaptation based on online performance evaluation for post-stroke subjects. In particular, the aim of the system is to implement the "assist-as-needed" paradigm based on actual patients' motor skills during a therapy session with an active upper-limb robotic exoskeleton. The strength of the work is to propose a real-time algorithm for the assistance tuning based on an "assistance-performance" relationship. Such a relationship is based on experimental measurements, and allows the algorithm to compute a straightforward calculation of the assistance required. Finally, an assessment phase will show how the system provides assistance based on the difficulties experienced from the subjects, also facilitating their adaptation during the task.