- 荒木諒介, 長谷川昂宏, 山内悠嗣, 山下隆義, 藤吉弘亘, 堂前幸康, 川西亮輔, 関真規人
- 日本ロボット学会誌, vol. 36, no. 8, pp. 559–566, 2018.
Accurate grasping of objects such as industrial parts and everyday necessities is an important task for industrial robots and living-support robots. Many methods have been proposed for grasp point detection for robots, some that utilize machine learning and some that do not. Recently, a grasp-point detection method using a 2-stage deep neural network has been proposed. Although the 2-stage deep neural network could detect the grasping point of no-learned objects, the computation cost would be high. In this paper, we propose a method for detecting grasping points using one deep convolutional neural network (DCNN) introducing graspability. Simultaneous detection of grasping points and graspability in one neural network lessens calculation costs. Evaluation experiments confirmed that grasping points could be properly detected using graspability.