学習制御で得られたロボットの入力トルクパターンの非線形時間軸変換

抄録

To realize speedy and accurate motions of robot manipulators, it is well known that feedforward inputs are important. One of the method to realize the feedforward inputs is “Learning Control”. In such learning schemes, an ideal input torque pattern is formed through several actual iterative operations, only using input torque and output motion patterns of the previous trial. Therefore, the learning schemes can realize a desired motion easily. However, if we need many different motions, it is difficult to make the input torque patterns in the learning process, because the time and memory space to obtain and store the ideal input patterns are necessary. To overcome such difficulties, we propose a time scale translation for input torque patterns. Since translation of time scale means change of speed patterns of robot motions with a same spatial trajectory without using learning process. In the proposed translation, an arbitrary speed pattern is constructed only using four input patterns to realize four desired motions. Through some experimental results, the effectiveness of the proposed translation is demonstrated.