In-Rack Test Tube Pose Estimation Using RGB-D Data

Abstract

Accurate robotic manipulation of test tubes is pivotal in biology and medical industries to mitigate workforce shortages and enhance worker safety. A critical step toward successful manipulation is the accurate detection and localization of test tubes. In this paper, we present a three-staged framework to detect and estimate poses for the in-rack test tubes using color and depth data. The framework first employs a YOLO object detector to classify and localize test tubes and tube racks from image data. Subsequently, the tube rack’s pose is estimated through point cloud registration techniques. Finally, given the rack’s pose, we utilize an optimization-based algorithm with geometry constraints of the rack slots to determine the poses of individual test tubes. This strategic approach ensures robust pose estimation even when confronted with noisy or incomplete point cloud data.