フィードバック誤差学習によるロボット・マニピュレータのファジィ制御

抄録

Feedback control is inadequate for the fast control of robotic manipulator with low reduction ratios between joints and actuators, since the nonlinearity of manipulator dynamics is by no means negligible.
Hierarchical neural network models which account for the generation of motor command of the manipulator whose dynamics are not known, have been proposed recently. In this paper a new method of learning control of manipulator using Sugeno's fuzzy reasoning is proposed, and compared with the ones using neural network models by computer simulations.
It is shown that the multi-layered neural network model possesses a great ability to generalize learning, but substantially long period is required for repetitional learning of a short movement pattern. The fuzzy control proposed in this paper does not have enough ability to generalize learning, but is much superior in the speed of learning and does not require the model of the manipulator dynamics.