Combining Wrist and Hand Gestures to Increase the Number of Classes in Myoelectric Control.

Replicating coordinated and natural multi-joint movements in myoelectric prosthetics remains challenging due to the complexity of controlling multiple degrees of freedom (DoFs). This study explores the formation of a distinct "combination class" by merging dual gesture classes, such as hand gestures and wrist movements, while maximizing their separability to improve the classification. Surface electromyography data were recorded in 9 subjects to assess whether these combined movements can be reliably classified as independent classes with respect to their component movements (individual gestures), using collected as well as artificially reconstructed data to reduce training time. The separability of combination classes within the feature space was assessed using Bhattacharyya distances (BD), while the classification performance was evaluated using a Linear Discriminant Analysis (LDA) classifier. The results showed that combination classes formed distinct regions in the feature space and that they can be classified with a high success rate (97%) when using collected data. However, when using the artificially synthesized data, the classification performance exhibited a 17% decline, underscoring the challenges of computationally replicating multi-DoF dynamics. These findings highlight the feasibility of combining highly separable hand and wrist movements to generate new classes for multi-DoF control and the limitations of synthesized data in accurately replicating these combinations.