This research concerns a new method for the learning control of robotic manipulators based on self-tuning of computed torque gains using an artificial neural network (ANN). We consider some ways to use ANNs in learning control schemes so that ANNs become simplified for multi-input/output systems. We discuss two important issues in ANNs : the first concerns a sigmoid unit function used to achieve flexibility in ANN structure and the second is a learning algorithm for the sigmoid function that is similar to the popular back-propagation algorithms. It is shown that this algorithm has an excellent generalization property which is superior to back propagation in error minimization capability. Two kinds of sigmoid functions are most widely studied and applied to neural network structure today : one is bipolar (hyperbolic tangent function) with a range from -1 to 1 and the other is unipolar with a range from 0 to 1. However, in our proposed neural network structure, the sigmoid unit functions are changeable in shape depending on the training data. The proposed neural network structure has more than one linear unit at the input layer, only one bipolar unit with changeable shape at the hidden layer, and more than one unipolar unit with changeable shape at the output layer. Such a structure, in comparison with conventional structures, becomes very simple and can be utilized in large-scale elaborate systems. A simulation is demonstrated to evaluate the newly proposed structure by applying the method to construct an adaptive computed torque controller for a two-link manipulator.
[1]
Takayuki Yamada,et al.
Neural network controller using autotuning method for nonlinear functions
,
1992,
IEEE Trans. Neural Networks.
[2]
S. Omatu,et al.
A neural network controller for a temperature control system
,
1992
.
[3]
Richard S. Sutton,et al.
Neural networks for control
,
1990
.
[4]
Toshio Fukuda,et al.
Mobile robot control using fuzzy-Gaussian neural networks
,
1993,
Proceedings of 1993 IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS '93).
[5]
P.J. Antsaklis,et al.
Implementations of learning control systems using neural networks
,
1992,
IEEE Control Systems.
[6]
A. Sideris,et al.
A multilayered neural network controller
,
1988,
IEEE Control Systems Magazine.
[7]
Toshio Fukuda,et al.
Force control of robot manipulator by neural network. (Control of one degree-of-freedom manipulator).
,
1989
.