A life-like control algorithm for the natural movement of a 3 dof finger

This paper describes the dynamic control of a 3 dof finger emulating a human medium finger for reaching a desired configuration in space. The controller imitates the physiological control and the motion planning strategy in human beings. A Multi Layers Perceptron (MLP) architecture has been setup, which receives as input the desired position and velocity of the fingertip (sensorial input) and the actual position and velocity (proprioceptive input). The output produces the necessary torques to reach the desired position. The MLP is fed back by the plant with the actual positions and velocities by closing the loop with a direct dynamics block: the movement due to the applied torques is sent back to update the proprioceptive input. Assuming a ´cognitive´ approach well established in the literature, the controller performs a smooth ´natural´ speed profile of the fingertip motion by adopting an arranged extension of the minimum jerk theory. The generalization capabilities of the NN is checked. The results are discussed by considering the closed loop configuration with respect to the open loop one (that is the MLP without the feedback). We suggest this approach for dynamical control of multi-finger prosthetic/robotic hand.