Control of dynamic grasping systems using neural network approximation

Abstract

A control algorithm for dynamic grasping systems using neural network approximation (NNA) is proposed. The kinematic and dynamic equations of the grasping system are derived. Based on these equations, a method for generalized computed torque control is developed. From computations of this control scheme, four elemental operation functions that are realized by elemental NNA functions are induced. All of the control computations in the grasping system are accomplished using neural network approximation. The PD control of a two-jointed finger mechanism is studied as an example of the application of the algorithm. Results using the NNA functions are compared.<>