An important characteristic of the human hand is the mechanical advantage, called as the moment arm, that each muscle-tendon combine has on each joint. The relationship between the muscle-tendon excursions and joint movements depends on the bone shapes over which tendon travels and structure of the tendon network, and is characterized by the moment arm. In case of the index finger a moment arm matrix can be defined to relate excursions of seven muscles to four independent joint motions. From cadaver and human studies it is known that the moment arm matrix is nonlinear function of all joint angles and that is varies significantly from a person to another. Since the moment arm matrix relates the muscle excursions to the joint motions and also the muscle forces to the joint torques, understanding the variations of the moment arm matrix is critical for understanding the hand control strategies and hand dexterity.

Our idea is to gain an understanding of the index finger moment arms by determining and analyzing moment arm matrix for ACT Hand index finger. This project presents a system identification technique that allows, for the first time, the determination of the moment arm matrix as a function of multiple joints moving simultaneously. For the moment arm determination for this higher dimension problem, we employ least squares regression and a machine learning technique called Gaussian Processes. We compare the results from the two techniques, validate our results with test data and also compare our results with the available cadaver data. We present implications of determination of variable moment arm matrix toward understanding the biomechanical properties of human hand and for the neuromuscular control for the ACT Hand index finger motion.