VAE-Based Synthetic EMG Generation with Mix-Consistency Loss for Recognizing Unseen Motion Combinations

Recent advancements in electromyogram (EMG) signal processing have led to a novel method for generating synthetic motion patterns, crucial for enhancing prosthesis control. Traditional approaches often relied on the assumption that combined motions could be represented as linear combinations of basic movements, a premise that frequently fails due to complex neuromuscular interactions. The newly proposed method utilizes a variational autoencoder (VAE) to encode EMG signals into a structured latent space, allowing for the synthesis of more accurate combined motion patterns. By introducing a mix-consistency loss, the method ensures that these synthetic patterns are effectively embedded between their constituent basic motions. This innovative approach has demonstrated a remarkable 30% improvement in classification accuracy during upper-limb motion classification experiments involving healthy participants. The implications of this research are significant, as it not only enhances the perfor…