2022 Fiscal Year Final Research Report
Feature extraction from vibrotactile data using autoencoder
| Project/Area Number |
21K18680
|
|---|
| Research Category |
Grant-in-Aid for Challenging Research (Exploratory)
|
| Allocation Type | Multi-year Fund |
|---|
| Review Section |
Medium-sized Section 18:Mechanics of materials, production engineering, design engineering, and related fields
|
|---|
| Research Institution | Keio University |
|---|
Principal Investigator |
|
|---|
| Project Period (FY) |
2021-07-09 – 2023-03-31
|
| Keywords | 触感センシング / 機械学習 / オートエンコーダ / 畳み込みニューラルネットワーク / ウェーブレット変換 |
|---|
| Outline of Final Research Achievements |
Vibration information was obtained when a tactile sensor traced on a sample. Then, features were extracted from the obtained vibration data using the deep autoencoder. Next, we obtained tactile scores when a subject touched the same samples in a sensory evaluation experiment with a seven-scale semantic differential method, followed by constructing a tactile estimation model with the acquired features as input and the sensory evaluation scores as output. In addition to this model, feature vectors were extracted from the vibration data by using a convolutional neural network. In this case, we introduced wavelet transformation in order to obtain a scalogram, two-dimensional image, of the vibration data. Note that a convolutional neural network is suitable for extracting features from two-dimensional image. As a result, it is shown that the tactile score can be predicted with an error of less than one point from the average value of sensory evaluation.
|
| Free Research Field |
機械工学
|
| Academic Significance and Societal Importance of the Research Achievements |
本研究では，機械学習を援用することにより，触感センサから得られた機械的データに適した特徴量の抽出が触感推定に有効であることを示した．これにより，多くの研究では見逃されていた振動データに内在した潜在的な特徴量を顕在化でき，触感研究を変革する可能性を秘めている．すなわち，1990年代から長く研究され，触感研究の根幹を成すにも関わらず，未だに確立されていない定量的触感計測を可能にする意義は大きい．また，社会実装が加速する機械学習の適用範囲拡大の挑戦としても意義深い．すなわち，本研究の成果は画像データに広くて適用される機械学習を触感データへの適用に拡張する可能性を示している．
|
