Adaptation and learning for robotic manipulator by neural network
Toshio Fukuda, Takanori Shibata, Masatoshi Tokita, Toyokazu Mitsuoka · 1990
Neural network applications for robotic motion control in which the controller is applicable to position and force control of robotic manipulators are addressed. The proposed neural servo controller is based on a neural network which consists of input/output layers and two hidden layers, and which has time delay elements in its first hidden layer. This neural network can learn the complex dynamics of the system in forward manner to cooperate with the feedback loop, depending on the unknown characteristics of objects to be handled. A variable learning method, fuzzy turbo, which is based on fuzzy set theory, is proposed. This method can avoid stagnation during the learning process and has insensitive characteristics at a stable extreme, so that the neural network can learn the dynamical system quickly. Simulations are carried out for the case of force control handling of unknown objects and trajectory control handling of unknown payloads of a two-dimensional robotic manipulator.>