Development of Artificial Intelligent Techniques for Manipulator Position Control

Abstract

Inspired by works in soft computing this research applies the constituents of soft computing to act as the "brain" that controls the positioning process of a robot manipulator's tool. This work combines three methods in artificial intelligence: fuzzy rules, neural networks, and genetic algorithm to form the soft computing plant uniquely planned for a six degree-of-freedom serial manipulator. The forward kinematics of the manipulator is made as the feedforward control plant while the soft computing plant replaces the inverse kinematics in the feedback loop. Fine manipulator positioning is first achieved from the learning stage, and later execution through forward kinematics after the soft computing plant proposes inputs and the iterations. It is shown experimentally that the technique proposed is capable of producing results with very low errors. Experiment A for example resulted the position errors onpx: 0.004%;py: 0.006%; andpz: 0.002%.