Deep Learning for Assistive Decision-Making in Robot-Aided Rehabilitation Therapy

The use of robotic rehabilitation devices has emerged as a promising approach to enhance motor recovery during rehabilitation. One of the significant challenges while using these devices is the ability to decide when to provide assistance to the patient. In this regard, we propose a Deep Learning-based solution that can learn from a therapist’s criteria when a patient requires assistance during robot-aided rehabilitation therapy. We have trained and evaluated the proposed model with data from different patients during a point-to-point game modality. To make the model more universal, we have applied a series of transformations to the trajectory data before using them as inputs. The proposed model has been evaluated using different metrics and has shown an accuracy of 93.21% and an F1-Score of 85.05% with the validation dataset. Furthermore, the model has achieved an accuracy of 69.32% and an F1-Score of 63.31% with users who were not involved in the model learning process.