2019 Fiscal Year Annual Research Report
Dense 3-axis tactile sensing and AI to implement human-like manual skills in robots
| Project/Area Number |
19H02116
|
|---|
| Research Institution | Waseda University |
|---|
Principal Investigator |
シュミッツ アレクサンダー 早稲田大学, 理工学術院, 准教授(任期付) (30729455)
|
|---|
| Co-Investigator(Kenkyū-buntansha) |
尾形 哲也 早稲田大学, 理工学術院, 教授 (00318768)
玉城 絵美 早稲田大学, 理工学術院, 准教授(任期付) (30515086)
Somlor Sophon 早稲田大学, 理工学術院総合研究所(理工学研究所), 次席研究員(研究院講師) (40791231)
|
|---|
| Project Period (FY) |
2019-04-01 – 2022-03-31
|
| Keywords | Tactile Sensing |
|---|
| Outline of Annual Research Achievements |
In this research we develop a smart sensing system that enables robot hands to achieve human-like manipulation skills. Key components are 1. dense 3-axis tactile sensors for robot hands and 2. learning algorithms exploiting massive 3-axis tactile data for intelligent force control.
We integrated the tactile skin sensors in grippers and robot hands. Using a novel joint (with a remote center of motion mechanism) we could achieve full coverage of the palmar side of the fingers with sensors in one gripper. Furthermore, we instrumented human hands with the sensors, to enable skill transfer from human to robot hands in the future. We used the skin sensors integrated in the robot hands for various machine learning experiments. In particular, we used deep convolutional neural networks for tactile object recognition as well as for in-hand manipulation.
|
| Current Status of Research Progress |
Current Status of Research Progress
1: Research has progressed more than it was originally planned.
Reason
Overall our progress this year was better than expected. We already performed more machine learning experiments than expected.
|
| Strategy for Future Research Activity |
We will further work on the integration of the sensors into various robot hands and arms. We will continue to use the sensors for various machine learning experiments. We are dealing with an unprecedented amount of force vector measurements. Convolutional deep neural networks are the state of the art for visual data and we have shown that they can be used for tactile data as well. Indeed, the matrix-like 3D force data resembles RGB 3-dimensional visual data. We will continue to use such machine learning methods.
|
Research Products
(14 results)
