ロボット動作計画のための３次元点群を用いた実配線の変形予測

抄録

To meet the demand of mass customization and address labor shortages in industry, the implementation of flexible production line with industrial robots is essential. Reconfiguring production lines requires robot motion planning, but discrepancies between the simulated environment and operational environment—such as position or shape errors—often leads to rework. This study aims to prevent rework in robot motion planning during production line reconfiguration, especially focusing on the deformable cable fixed across a robot joint. The proposed approach integrates the robot's operational environment, including the deformable cable, into the simulated environment using 3D measurements. The deformable cable is extracted through point cloud registration, and its deformation is predicted by introducing angular and radial deformations in polar coordinates around the robot joint axis. When registering point clouds with discrepancies between the operational environment and simulated environment, the iterative closest point (ICP) method with threshold is effective. However, determining the appropriate threshold value usually involves trial and error. To address this issue, this study proposed threshold determination methods that account for discrepancies in part shape, part existence and measurement errors. To predict cable deformation extracted through ICP with the determined threshold, a method based on the deformation curve of a fixed-ended beam is introduced. This eliminates the need for manual parameter adjustment in physical simulations through trial and error. Validation in a real-world robot environment showed that proposed ICP threshold achieved the lowest registration error, ranging from 1 to 100mm. Furthermore, cable deformation prediction yielded a maximum error of 2.9 mm within joint angles of 0 to 90 degrees. An application test demonstrated a 92% reduction in rework during robot motion planning. This study contributes to the development of flexible production lines with industrial robots by providing accurate deformation predictions without relying on physical simulations.