学習終端制御法および弾性ロボットアームの制御への応用

概要

論文の詳細を見る
The control input which moves a linear discrete-time system to goal states in finite time can be calculated in advance if the dynamics of the system are well understood. However, when the identified model is different from the actual system, residual state errors at the end of the motion are unavoidable. In this paper, a control scheme which improves the accuracy of the terminal-point control of the discrete-time system by learning is proposed, and the convergence of the learning process is discussed. Moveover, in the case that there are constraints on control inputs, a learning terminal-point control in which a weighted pseudoinverse matrix is utilized is presented. These control methods are applied to a flexible robot arm, and it is shown through experiments that they are able to realize high-speed motions and suppress residual vibrations.
社団法人日本機械学会の論文
1992-09-25