運動パターン学習による2足歩行ロボットの歩行実現

概要

論文の詳細を見る
In biped locomotions there are several kinds of walking motion patterns. By selecting one of these motion patterns appropriately according to the change of circumstances, a robot may be afforded abilities of adaptivity in walking. However, if there is no large change of circumstances around the robot, the biped locomotion can be realized by repeating the same motion. Therefore, first of all it is most necessary for the biped locomotion robot to generate one walking motion pattern.<BR>In general, even if one walking motion pattern is given, it can not be realized by only simple servo-mechanisms, since dynamic interferences of links of a robot, frictions, and changes of a mechanical constraint condition with walking motion strongly affect the motion of the robot. To overcome these difficulties we apply a kind of learning control to the biped locomotion robot. According to this control scheme, a desired input pattern which can realize a walking motion is constructed as the robot repeats walking motion. After several trials, the robot can obtain the walking motion pattern.<BR>In this paper, some experimental results obtained by this learning control are shown. Moreover, we suggest that an input pattern which generates another walking motion can be directly constructed from the input pattern obtained previously by learning.
一般社団法人日本ロボット学会の論文