見た目に影響が小さいマーカによる物体位置姿勢推定

抄録

We aim to develop stocking and disposal robots without changing the impression of products, and deal with a new problem of estimating products’ poses using appearance-free markers. To solve the problem, we propose to define the appearance-free marker as the marker which human cannot find within 3 sec, based on the psychology. However, it is difficult for robots to find appearance-free marker. Therefore, we develop a new pose estimation method which uses the appearance-free marker with making the robot’s camera approach the products based on the rough poses estimated by the deep learning. We performed impression evaluations. 100 participants’ evaluation showed that the size of appearance-free marker was 5 mm. Moreover, we evaluated our method by using our arm robot PA10. We implemented our method into PA10 which made its camera approach the products to estimate the products’ poses as accurately as the stocking and disposal robot which won the world competition by using 45 mm markers did. Experimental results showed that PA10 could estimate products’ poses with position errors of 11.9 mm and posture errors of 4.23 degree, and could stock and dispose products.