A Computational Method to Identify Optimal Functional Muscle Synergies from Estimated Muscle Activations

Achieving reliable muscle synergies (MSs) is critical for optimizing motor control in neuromusculoskeletal robotics. Existing MS extraction approaches are limited by traditional electromyography (EMG) acquisition constraints, variability due to methodological differences, and difficulty in assessing the contribution of deep and small muscle groups to optimal MSs. To overcome these challenges, we developed the determination of optimal functional MSs from estimated muscle activations (DOMSn-EMAs), a novel computational approach elucidating the impact of high-density estimated muscle activations (EMAs) on optimal functional MSs under specific biomechanical task constraints. DOMSn-EMAs comprises three key steps: 1) computing high-density EMAs through measured human motion data and high-fidelity musculoskeletal modeling with a modified computed muscle control (CMC) algorithm, constrained by limited surface EMG (sEMG) tracking; 2) extracting sparse MS properties using sparse nonnegative matrix factorization (SNMF) to improve reconstruction accuracy and repeatability; and 3) introducing a composite reliability index for optimal MS number identification. Validation on a publicly available dataset from ten subjects confirms DOMSn-EMAs' reliability. The results identify five optimal MSs from 43 EMAs, with an independent MS involving deep and small muscle groups during the early swing phase. This approach corroborates the similarity between EMG-derived and EMAs-derived synergies while identifying EMAs' potential influence on optimal functional MSs. DOMSn-EMAs provides a pioneering and practical approach for synergy analysis, ensuring more reliable and sophisticated MSs than previous modular architectures for robotics applications.