Summary

Recently, there has been a lot of interest in applying fuzzy logic or/and neural networks to the control of nonlinear systems such as robot manipulators. It is known that fuzzy control [¹–³] is able to deal with orally expressed human knowledge, because fuzzy control rules can express the vagueness. Consequently, the precise mathematical models of the plant and the environment are not needed to design the controller. It is possible to construct effective nonlinear controllers for the control of complex systems using fuzzy control techniques. There are some difficulties, however, in designing the fuzzy controller systematically. Furthermore, once fuzzy control rules and membership functions are designed, usually they are not modified, even if the controller is not very effective. On the other hand, it is known that neural network controllers [⁴–⁸] are able to learn from their experiments and adapt to a new environment online. Therefore, neural network control is especially effective when dynamics of the robot and/or the environment are uncertain because of its adaptation ability. However, it is difficult to design an effective neural network controller without prelearning and to understand the meaning of each weight value of trained neural networks. Besides, it is impossible even for a neural network controller to adapt to any kind of environment immediately. Consequently, although an adaptive neural network controller is one of the most effective controllers of applied force to an unknown environment by a robot manipulator, the robot manipulator might cause some damage to the environment before the neural network adapts to it. To cope with these problems, fuzzy neural networks, a combination of fuzzy logic and neural networks, have been applied to control robot manipulators, to combine both fuzzy control and neural network control advantages [⁹–¹⁴]. The fuzzy neural network is achieved by designing fuzzy control rules, using expert knowledge, and then modeling them as a neural network to create a learning ability.