LEARNING OPTIMIZATION FOR CPN-BASED TRAINING IN ROBOT POSITIONING CONTROL

Artificial neural net (ANN) models have been applied to the inverse kinematic problem for controlling robot positions. The selection of ANN training parameters, however, is an important vet complicated step which has to be taken before an ANN model for robot positioning control can be implemented effectively. The objective of this research is to utilize the counterpropagation network (CPN) for inverse kinematic mapping and obtain the best performance possible by systematic adjustment of network parameters. Taguchi statistical methods, efficient methods for analyzing the capability and accuracy of a system, have been used in this study. The working envelope of the robot simulated in this research is 150 x 150 x 60 mm3. The optimal accuracy and standard deviation determined by this research are 2.62 mm and 1.2 mm, respectively.