New Heuristic Method Merging Artificial Vision and Neural Networks used in a Sensorless Robotic Arm Position Control

Abstract

Inspired by the control system of voluntary movements developed in the human body based on vision and neural system, this paper presents a new heuristic method merging artificial vision and neural networks used in a sensorless robotic arm position control. This proposal is based on a structure of six artificial neural networks (ANN) of perceptrons, which correct the position of the arm in one of the six predefined directions, four in a the projection plane (forward, backward, right and left) and two in the vertical plane (up and down). The robotic arm displacement is based on the choose performed by the ANN processing the images capture by a camera, thus the chosen of the corresponding direction is derived from knowledge obtained during the supervised learning using similar situations. Finally, experimental results of the ANN learning process and robotic arm positioning tests are presented.