To realize desired motions, feedforward inputs play an important role. One way to form the feedforward input is
“Learning Control” based on actual iterative operations. Disadvantages of such learning schemes are summarized as follows. (1) For each desired trajectory we need several trials. It may take long time. (2) To memorize many ideal input patterns a large amount of memory space in a digital computer is necessary.
To overcome the above difficulties, we propose interpolation methods in this paper. In the proposed methods, several ideal input patterns are obtained through the learning control and stored in the memory at first. Next, by interpolating acquired ideal input patterns, another ideal input pattern is formed without iterative operation of the learning control.
We explain one method which is applicable to the case that the spatial trajectories of each desired motion are the same and the time trajectories are different. For the case that the spatial trajectories are different but has a linear relation in a task-oriented coordinates, the other interpolation method is mentioned in order to obtain approximate in-put patterns. In experiments that a robot moves along a conveyor with different speeds or different spatial trajectories, the effectiveness of the proposed methods is demonstrated.
抄録全体を表示