Neural networks for multi-finger robot hand control

Abstract EN

This paper investigates the employment of Artificial Neural Networks (ANN) for a multi-finger robot hand manipulation in which the object motion is defined in task-space with respect to six Cartesian based coordinates.

The approach followed here is to let an ANN learn the nonlinear functional relating the entire hand joints positions and displacements to object displacement.

This is done by considering the inverse hand Jacobian, in addition to the interaction between hand fingers and the object being grasped and manipulated.

The developed network has been trained for several object training patterns and postures within a Cartesian based palm dimension.

The paper demonstrates the proposed algorithm for a four fingered robot hand, where inverse hand Jacobian plays an important role in robot hand dynamic control.