ANN Based Robotic Arm Visual Servoing Nonlinear System

Abstract

Abstract The paper presents an ANN-based mechanism to model inter-related visual kinematics relations, which are an important fragment of visual servo closed loop system. The primary goal is to visually servo a 6-DOF robot arm, from an initial location toward a desired posture using only image data from a scene provided during the arm maneuver. Arm movements, dynamics, and kinematics were simulated with Matlab Robotics Tool Box. The arm is equipped with a CCD camera, where the visual CCD part is modeled and simulated with Matlab Epipolar Geometry Toolbox. The main issue that hinders visual servo system is related to the variant feature Jacobian matrix. The methodology used is based on the concept of integrating a multilayer ANN with an Image Based Visual Servoing system. ANN was used to learn, approximate, and map the highly complex relations that relate a scene movements to an arm movement through a visual servo. The proposed technique has proven to be an effective approach, that has resulted in reducing computational time.