ANN control based on patterns recognition for a robotic hand under different load conditions

Abstract EN

In this paper, the Artificial Neural Network (ANN) is trained on the patterns of the normal component to tangential component ratios at the time of slippage occurrence, so that it can be able to distinguish the slippage occurrence under different type of load (quasi-static and dynamic loads), and then generates a feedback signal used as an input signal to run the actuator.

This process is executed without the need for any information about the characteristics of the grasped object, such as weight, surface texture, shape, coefficient of the friction and the type of the load exerted on the grasped object.

For fulfillment this approach, a new fingertip design has been proposed in order to detect the slippage in multi-direction between the grasped object and the artificial fingertips.

This design is composed of two under-actuated fingers with an actuation system which includes flexible parts (compressive springs).

These springs operate as a compensator for the grasping force at the time of slippage occurrence in spite of the actuator is in stopped situation.

The contact force component ratios can be calculated via a conventional sensor (Flexiforce sensor) after processed the force data using Matlab/Simulink program through a specific mathematical model which is derived according to the mechanism of the artificial finger.