ビンピッキングのための環境情報を用いた物体検出と動作計画との連携

Abstract

This proposed bin-picking method removes the inconsistency in the information transferred from object detection to path planning of manipulator by using environmental information. The inconsistency is that the object detection only provides the position and orientation of a few detected objects while the path planning needs the environmental information including all objects in the work field. To solve this problem, we propose to include possible objects representing undetected objects in the environmental information. A Hough transform, which records the model-to-image matches in a parameter space, is used to estimate the pose. Both detected and possible objects are extracted from the parameter space. To reduce the number of detection errors and the object-detection time, a pair of object features that reduces the number of invalid votes in the parameter space is used for the Hough transform. Simulated path planning and an experiment on casting objects using a range finder and a manipulator showed that the environmental information including the possible objects improves the ability to plan a safe path for the manipulator. Simulated object detection and an experiment showed that using a pair of features makes the process faster and reduces the number of invalid votes.