The goal of this project is to improve robotic neurorehabilitation by developing robotic artificial intelligence, which can automatically adapt the robotic training strategy to the trainee’s special needs and the trained motor task, and improve its strategy over time.
In the first part of the project, machine learning methods are used to identify which performance features (e.g., timing of movements, smoothness, etc.) lead to higher scores for a trainee, within the trainee’s natural performance variability. Robotic training intervention are then modified/adapted to reinforce the training of these identified features that lead to better general performance, in order to increase the ultimate task performance as a result. In the second part, reinforcement learning methods are used to cumulatively store and improve the “experience” of robot in a given task/game, in order to improve its decision efficiency about the selection/adjustment of the training.
These novel intelligent and adaptive robotic approaches could be applied to train a large range of different motor task, such as activities to daily living, and could be employed in the robotic devices already in use in the clinics. Therefore, our approach has the potential to result in more effective and efficient neurorehabilitation training approaches that would enhance motor recovery in neurologic patients.