I. Introduction
Among the major factors contributing to the dexterity and versatility of the hand is the uniqueness of the thumb. It is widely claimed by clinicians that more than 50 percent of all hand functions are made possible by the thumb [1]. It naturally follows that any damage to the hand and more specifically to the thumb will have a negative impact on human functioning. For this reason many researchers have invested heavily in producing artificial limbs to be used in cases where there is a complete or partial loss of hand. One of the most successful prosthetics hands created in recent times has been PRODIGITS made by Touch Bionics in Scotland [2]. This was developed further into the i-Limb Ultra revolution which is a more advanced prosthetic with features like lighter and more anatomically accurate fingers, powered thumb rotation, added dexterity (up to 24 different grip patterns) and even Bluetooth connectivity [3]. Another such advanced prosthetic is the Bebionic3 by [4], that has 14 pre-programmed grip combinations for the most common hand movements like typing, precise handling and so on. It is controlled by the muscle signals generated by the upper arm. However, as apparent from a survey of the currently available prosthetics, most of the development in this area is based on discontinuous or discrete thumb position control, which makes it possible to recreate many of the most important functions/positions of the thumb but does not result in total imitation. These prosthetics are designed to emulate the movement, design and performance of a natural hand as closely as possible in order to restore the lost functionality. It is therefore important that an artificial limbs be developed that can mimic the actions of a real body part as closely as possible. There have been recent studies in this particular area, such as the one by [5] who suggested the use of biomechanical models to predict the force at the thumb-tip based on the sEMG signals. For this they used the Hill-based muscle model, which uses the processed neural activity, muscle length and contraction velocity to estimate the force generated by the muscle. This model was applied to five of the nine muscles (the remaining four muscle forces were estimated) and the resultant force at the thumb-tip was estimated. Similarly, [6] have suggested a way for continuous position control of the wrist joint using a model-free technique. They use a multi-layer neural network, which is trained through back propagation to derive a time-continuous relationship between the EMG signals and the wrist angle, which was then used to control a 1-Degree of Freedom manipulator.