2A1-B10 ニューラルネットによる腱駆動ロボットアームの制御 : ランダムな運動からの逆モデルの学習(移動知)

概要

論文の詳細を見る
In order to conduct the long time exploration-based learning experiments with a physical robot, there was necessity to build a durable robot. Therefore, we built a novel robot with a tendon-based actuation mechanism which can afford elasticity at each joint of the robot. Each of the arm joints is actuated by an antagonistic pair of two motor-spring assemblies. By controlling the tension in the spring with a classic PID controller, the joints are being torque-controlled. Because of inherent nonlinear characteristics of the developed system, standard PID control schemas cannot be applied to the system. Therefore, we employ the RNN to construct the inverse model which plays the role of controller. In this paper, we report the result of preliminary experiment using this controller.
一般社団法人日本機械学会の論文
2007-05-11