Title :
A self-organized CMAC controller
Author :
Chow, Mo-Yuen ; Menozzi, Alberico
Author_Institution :
Dept. of Electr. & Comput. Eng., North Carolina State Univ., Raleigh, NC, USA
Abstract :
A neural network structure that has been particularly successful in robotic control is the cerebellar model articulation controller (CMAC). In this paper, a CMAC network controller that uses self-organization through competitive learning is presented. The concept consists in applying the self-organizing characteristic of a Kohonen map to a CMAC neural network. This allows the CMAC network to organize its neurons efficiently. The approach can be applied on a simple two-link robot arm model which approximates the agonist-antagonist activity of muscles in the human arm. The CMAC and SOCMAC controllers can be trained to learn the behavior of a conventional PI controller, and comparative results are presented. The training procedures and the parameters that are involved in the design of the self-organizing CMAC (SOCMAC) controller are discussed
Keywords :
cerebellar model arithmetic computers; intelligent control; manipulators; neurocontrollers; self-organising feature maps; unsupervised learning; Kohonen map; cerebellar model articulation controller; competitive learning; neural network; self-organization; self-organized CMAC controller; two-link robot arm model; Artificial neural networks; Biological neural networks; Feedforward neural networks; Humans; Mathematical model; Multi-layer neural network; Neural networks; Neurons; Robot control; USA Councils;
Conference_Titel :
Industrial Technology, 1994., Proceedings of the IEEE International Conference on
Conference_Location :
Guangzhou
Print_ISBN :
0-7803-1978-8
DOI :
10.1109/ICIT.1994.467181