ランダムフォレストとベイズ最適化を用いた大型産業用ロボットの位置決め補正法

抄録

In pipe manufacturing, a technology has been developed to simulate two flanged surfaces connected by a pipe using two industrial robots. This system enables us to check whether the pipes can be connected without any problems in the factory instead of in the field. Because of this system, it is not necessary to go back and forth between the factory and the site as in the conventional system, and time and cost can be reduced. However, for this method to be effective, a high degree of positioning accuracy of the robot is required. Therefore, in this paper, we study the motion accuracy of a robot and its compensation method for constructing this system using an industrial robot. In this experiment, we use a laser tracker to measure the robot’s end effector. At the same time, the joint angles of each joint, the maximum motor current and other servo information were measured. The random forest method, a kind of machine learning technique, is used to predict the positional error. As a result, one of the causes of the positioning error is identified to be the direction of rotation of the robot's pointer. Next, using the Bayesian optimization and the predicted positional error by the random forest method, we searched for the coordinate with the smallest positional error and fed back the resultant positional error. As a result, it is confirmed that the proposed method makes it feasible to reduce the positioning error with an offline teaching.