表題番号：2023C-496
日付：2024/02/05
研究課題ヒトとロボットの協調作業における物体形状の変形加工の再現機能に関する研究
| 研究者所属（当時） | 資格 | 氏名 | |
|---|---|---|---|
| （代表者） | 理工学術院 大学院情報生産システム研究科 | 教授 | 松丸 隆文 |
| （連携研究者） | Graduate School of Information, Production and System, Waseda University | Doctoral student | Xin He |
| （連携研究者） | Faculty of Mechatronics, Warsaw University of Technology | Vibekananda Dutta | Vibekananda Dutta |
| （連携研究者） | Faculty of Power and Aeronautical Engineering, Warsaw University of Technology | Professor | Teresa Zielinska |
- 研究成果概要
- This study examines a low-cost (without expensive datasets) general strategy for reproducing plastic deformation by robotic manipulation. Under this study, this work aims to achieve the Semi-3D (limited complexity and variations) deformation reproduction of plastic objects (reproducing various target trenches manually made in a box of kinetic sand with different structures, depth, and width variations by employing a 6DOF manipulator and an Azure Kinect RGB-D sensor). To achieve the goal, we contribute three main techniques:(1) a 3D surface points labeling method to classify the surface points according to the principal curvature so that can generate a smooth labeled result for each 3D point in an RGB-D image,(2) an improved deformation representation method based on the RANSAC (Random Sample Consensus) strategy for representing the various shapes in a standardized form so that can represent the shape by the combination of 3D splines and different variations,(3) a manipulation trajectory generation method for robotic systems based on a task-based adaptive periodic dynamic motion primitive.Different experimental settings were applied to assess the performance of our proposals. Through the comparison experiments (robotic manipulation vs. human operation), we calculated the Point Cloud Structural Similarity Metric (PointSSIM) and demonstrated deformation reproduction results with an acceptable similarity that close to the human operation (less than 3% of difference in average). Furthermore, our proposal can be further extended for more general (without constraints) 3D deformation reproduction. In addition, each proposed technique has the potential to be adapted in other different fields.For future work, our goal is to extend our deformation strategy to the general deformation (without constraints) of general 3D objects. The main challenge for the general 3D object will be the object’s splines (main structure) matching and deforming. If the main structure of 3D object target shapes can be represented from the initial state, extending the proposals of this study (new variations for general deformation cases and corresponding motion primitives for robotic systems), it is possible for a robotic system to reproduce the general deformation of general 3D objects.