Dimensionality Reduction in Control and Coordination of Human Hand.

The human hand is an excellent example of versatile architecture which can easily accomplish numerous tasks with very least effort possible. Researchers have been trying to analyze the complex architecture of the human hand. It is an unsolved mystery even today how Central Nervous System (CNS) controls the high degree of freedom (DoF) of the human hand. Investigators have put forth numerous theories which support movement planning both at higher and lower levels of the neural system as well as the bio mechanical system. This planning is hypothesized to happen in a reduced dimensionality space of tiny modules of movement called movement primitives often referred to as synergies. These synergies are physiologically significant in planning and control of movement. This dissertation presents time-varying kinematic synergies which linearly combine to generate the entire movement. The decomposition of these synergies becomes an exciting optimization problem and even more fascinating as it addresses two most important problems of motor control---coordination and dimensionality reduction. In this dissertation, a new model of convolutive mixtures for generation of joint movements is proposed. According to this model, an impulse originated in the higher-level neural system evokes the activation of some circuits in the lower-level neural system, then stimulates certain biomechanical structures, and eventually creates a stereotyped angular change at each finger-joint of the hand. Current model enabled greater access to existing blind source separation algorithms which reduce the computational complexity. First, kinematic synergies were extracted from a well known matrix factorization method, namely principal component analysis. By using the above kinematic synergies, a method to obtain temporal postural synergies is established. These temporal postural synergies were further used in the model of convolutive mixtures. An optimal selection of these temporal synergies which can reconstruct movements is then achieved by l1-minimization. The realization of the model by l1-minimization out performed the previous models which use steepest descent gradient methods. Synergies have received increased attention in the fields of robotics, human computer interface, telesurgery and rehabilitation. Improved performance and new computational model to decompose synergies presented here might enable them to be appropriate for real time applications.