A feedforward neural network controlling the movement of a 3-DOF finger

This paper describes the dynamic control of a 3 degree of freedom (DOF) ringer emulating a human finger for reaching a desired fingertip position in space. The control consists of a neural network (NN) which provides the necessary three torques to the phalanges granting a smooth "natural" speed profile of the fingertip motion. To obtain the results, we face these problems: 1) the elimination of the redundancy due to the third joint; 2) the mathematical description of a natural movement; 3) the calculation of the torques for executing movement; and 4) the optimization of the NN's structure. Assuming a "cognitive" approach well-established in the literature, the first and the second points are solved by adopting an extension of the minimum jerk theory. The classic Lagrange equations are applied to compute the three-motor torque. Finally, a multilayer perceptron (MLP) NN is trained to move the device in a natural manner. The generalization capabilities of the NN are checked on new never-seen movements, and different MLP architectures are compared on the basis of indexes representing the motor performance. The results suggest we should pursue this approach for multifinger hand in order to achieve a natural NN prosthetic/robotic dynamic control system.