CMAC AND HCMAC NEURAL NETWORK CONTROL OF 7-DOF REDUNTANT ROBOT KINEMATICS REDUNDANCY

Volume: 
Volume 07
Abstract 

The kinematics problems of redundant robots have been investigated for many years. Plenty of different applications for robot redundancy were implemented with success. Some of them were: improvement of redundant robot manipulability, robot obstacle avoidance, robot energy consumption optimization etc. Widely used methods use conventional approaches as for example in case of manipulability enhancement is used the gradient computation. However, the computational effort of these approaches brings many difficulties when it is used with other constraints. The solution to these problems is implementation of new intelligent methods based on artificial neural networks. This paper deals with application such methods where CMAC and HCMAC (Hierarchical Cerebellar Model Arithmetic Controller) neural networks were used in redundancy control of 7 DOF redundant manipulator. And manipulability enhancement constraint was chosen as redundancy constraint. First tested neural network was conventional CMAC neural network with supervised learning. It performed well in the terms of fast learning and local generalization capability. Nevertheless, the conventional CMAC has showed enormous memory requirement. Another tested neural network for the same task was HCMAC. It is shown that HCMAC perfectly approximated the testing function with relatively fast learning.

Author 
Vladislav Řikovský Štefan Kozák