Upper-Limb Synergy Modeling from Real-World Contactless Motion Tracking and Validation via Robotic Hand Rendering

Abstract

The extraction of low-dimensional kinematic synergies from movement data is foundational for understanding motor control and enabling intuitive human-robot interaction. However, conventional methods like Principal Component Analysis (PCA) are sensitive to the noise and artifacts inherent in practical motion capture. This study introduces Robust Principal Component Analysis (RPCA) as a superior alternative for synergy extraction from hand motion data. Analysis of hand motion data from 40 subjects across three cohorts (High, Medium, and Low-Quality) defined by landmark tracking quality revealed that PCA required 7-10 synergies to explain 95% of the variance, whereas RPCA achieved the same threshold with only 3-5 synergies, demonstrating a more compact representation. When dimensionality was fixed (k=4), RPCA consistently captured approximately 10% more variance than PCA and maintained stable performance as data quality degraded. Qualitatively, RPCA produced synergies with smoother temporal activation profiles and clearer separation between hand and arm dominant components, better reflecting known motor coordination patterns. RPCA also achieved lower normalized reconstruction error. The practical utility was validated by successfully mapping reconstructed postures to a 6-DoF anthropomorphic robotic hand. These results demonstrate that RPCA effectively isolates sparse tracking corruption to recover robust, low dimensional synergies, offering significant advantages for motor control analysis and the development of noise resilient human-machine interfaces.