Machine learning based Development and Evaluation of a Hand Exoskeleton Rehabilitation Robot for Patients with Partial Paralysis

This work focuses on the estimation of elbow joint angles from electromyography (EMG) signals and investigates the use of machine learning models in rehabilitation robotics. Four distinct machine learning algorithms were tested for their capacity to precisely translate EMG inputs into actuator commands: Artificial Neural Network (ANN), Naive Bayes (NB), Linear Regression (LR), and Decision Trees (DT). The models were trained and tested using a dataset generated by a wireless EMG sensor and motion capture system that recorded the dynamic relationship between elbow movements and related muscle activations. The results show the remarkable 98.7% accuracy of the ANN model, indicating that it can be a valuable tool for real-time rehabilitation robotics applications. Additionally, NB, LR, and DT displayed competitive results that demonstrated their unique capacities to manage a range of sensor inputs. The discussion highlights how important these models' interpretability, computational efficiency, and continuous adaptation are to their successful integration into real-world rehabilitation settings. This work contributes to the field of rehabilitation robotics by applying machine learning models for accurate actuation control based on EMG signals. The findings demonstrate how these models could enhance the accuracy and responsiveness of assistive technology, offering individuals with partial paralysis customized and adaptable rehabilitation regimens.