Efficient EMG Pattern Classification-based Control Interface for Exoskeleton Gloves Using Mahalanobis Distance to Handle Undefined Classes
By: Yuriah Lydon
Department: Engineering
Faculty Advisor: Dr. Xiaorong Zhang, Dr. Zhuwei Qin and Dr. David Quintero
The Electromyographic (EMG) signal has been established as a reliable bioelectrical signal for representing muscle activities. EMG pattern classification, which involves interpreting multi-channel EMG signals using machine learning techniques, is widely used to identify human movement intentions. However, one limitation of EMG pattern classification is its inability to handle undefined classes effectively, such as unintended motions or movement transitions that are not included in the training data, leading to misclassification and generation of incorrect control commands for external applications.
To address this challenge, our research aims to design a novel, low-cost, and efficient EMG pattern classification-based control interface for identifying two types of forearm movements to engage and disengage the grasping function of an exoskeleton glove (exo-glove). Our proposed control interface incorporates the Mahalanobis Distance (MD) algorithm, which enables the identification of data from undefined classes. This reduces the occurrence of unintended mis-engagement or disengagement of the exo-glove, thereby improving overall control accuracy.
An additional advantage of our control strategy is that it does not require collecting training data from undefined classes, which can be time-consuming for the user and require significant memory and computational resources for processing the training data. This makes the system more efficient and user-friendly in real-world applications.