学習によるマニピュレータの接触動作制御

―実機による検証―

抄録

In this paper, we present a contact motion control method between a robotic manipulator and a workpiece. A contact motion is a very short time phenomenon and the dynamics of the system is changed at the instance of collision. It is therefore difficult to control a contact motion with linear control theory.
In previous works, we have applied a learning control scheme to solve these problems for one or two degree of freedom manipulators and have shown some simulation results to verify feasibility of the method. With the proposed method, a dead-beat response of contact force has realized after some iterative operations.
In this paper, we discuss the feasibility of the proposed method from an experimental point of view. Firstly, we show a design indication of the settling time which causes a non-overshoot response with finite power. Secondly, we study the relationship between the number of observation and learning convergency in case of existing measuring noise. Thirdly, we discuss an influence of modeling error which is caused by compliance of gears or links of a manipulator. Some experimental results illustrate validity of our algorithm.