ビンピッキングのための高精度三次元計測とロバスト推定

抄録

Robotic bin picking of small industrial parts with metallic luster is a challenging task. Especially, the bin picking for metallic bolts with high success rate is very challenging, since the metallic bolts are part of in nearly all industrial products. High accuracy 3D measurement for metallic bolts is difficult due to specular and diffuse reflection caused by a screw part of bolts. To achieve high success rate for bin picking of metallic bolts, we propose the following three approaches. First, we show that 4 steps phase shift can achieve three-order accuracy for 3D measurement by mathematical analysis and adopt it. Second, we propose a novel robust and fast posture estimation algorithm named SAC-LMI (Sample Consensus with Linear Matrix Inequality). The proposed algorithm SAC-LMI can compute posture of the metallic bolts with high accuracy by using 3D measured point cloud having many outliers. Finally, we adopt a simple 1-DOF handling device called RAPiD for robust bin picking against calibration error for camera-robot posture. Considering the three approaches shown above, we have developed a robotic bin picking system which provides 95% success rates in industrial bin picking.