[go: up one dir, main page]
More Web Proxy on the site http://driver.im/
Next Article in Journal
The Use of Principal Component Analysis for Reduction in Sleep Quality and Quantity Data in Female Professional Soccer
Previous Article in Journal
Focusing on Cracks with Instance Normalization Wavelet Layer
You seem to have javascript disabled. Please note that many of the page functionalities won't work as expected without javascript enabled.
 
 
Font Type:
Arial Georgia Verdana
Font Size:
Aa Aa Aa
Line Spacing:
Column Width:
Background:
Article

Predicting Sensory and Affective Tactile Perception from Physical Parameters Obtained by Using a Biomimetic Multimodal Tactile Sensor

by
Toshiki Ikejima
1,2,*,
Koji Mizukoshi
1 and
Yoshimune Nonomura
2
1
POLA Chemical Industries, Inc., Yokohama 244-0812, Kanagawa, Japan
2
Department of Applied Chemistry, Chemical Engineering, and Biochemical Engineering, Graduate School of Science and Engineering, Yamagata University, Yonezawa 992-8510, Yamagata, Japan
*
Author to whom correspondence should be addressed.
Sensors 2025, 25(1), 147; https://doi.org/10.3390/s25010147
Submission received: 23 November 2024 / Revised: 23 December 2024 / Accepted: 28 December 2024 / Published: 30 December 2024
(This article belongs to the Section Wearables)

Abstract

:
Tactile perception plays a crucial role in the perception of products and consumer preferences. This perception process is structured in hierarchical layers comprising a sensory layer (soft and smooth) and an affective layer (comfort and luxury). In this study, we attempted to predict the evaluation score of sensory and affective tactile perceptions of materials using a biomimetic multimodal tactile sensor that mimics the active touch behavior of humans and measures physical parameters such as force, vibration, and temperature. We conducted sensory and affective descriptor evaluations on 32 materials, including cosmetics, textiles, and leather. Using the physical parameters obtained by the biomimetic multimodal tactile sensor as explanatory variables, we predicted the scores of the sensory and affective descriptors in 10 regression models. The bagging regressor demonstrated the best performance, achieving a coefficient of determination (R2) of >0.6 for fourteen of nineteen sensory and eight of twelve affective descriptors. The present model exhibited particularly high prediction accuracy for sensory descriptors such as “moist” and “elastic”, and for affective descriptors such as “pleasant” and “like”. These findings suggest a method to support efficient tactile design in product development across various industries by predicting tactile descriptor scores using physical parameters from a biomimetic tactile sensor.

1. Introduction

Tactile perceptions from touching materials are essential factors that influence the intention to buy or use a product [1,2,3,4,5,6]. These perceptions can be divided into two hierarchical layers: (1) the sensory layer, which involves the perception of physical parameters such as roughness and hardness, and (2) the affective layer, which reflects personal experiences and preferences, such as comfort and luxury [7,8,9]. Previously, tactile perceptions have been evaluated using vocabulary specifically developed to account for these hierarchical structures, providing subjective evaluation indices for product development [8,10]. Recently, we have developed a new tactile vocabulary that considers the hierarchical structure and suitability of descriptors for evaluating a wide range of materials [11]. This new tactile vocabulary enables evaluation across different material groups, although different vocabularies have been used in various industries. This improvement enables cross-industry comparisons of tactile perception, thereby facilitating the development of innovative tactile perceptions, e.g., luxury cosmetic products that mimic the tactile perception of silk fabrics.
In general, tactile evaluations involve having participants touch a material and evaluate it using specific terminology. These processes can be difficult to conduct in product development because of time and cost implications [12]. However, these complications could be significantly reduced if tactile evaluation scores were able to be predicted from physical parameters measured by devices. Several studies have examined the relationship between tactile perception and physical parameters. Chen et al. [13] evaluated 37 plates with different surface treatments using a 20-point scale for seven tactile descriptors: warm–cold, slippery–sticky, smooth–rough, hard–soft, bumpy–flat, wet–dry, and like–dislike. In addition, to calculate the correlation coefficient between tactile scores and physical parameters, they measured surface roughness, friction coefficients, compliance, and temperature using four instruments. Similar studies have examined textiles and confectionery packaging [14,15]. Kikegawa et al. [16] evaluated 11 tactile descriptors, including moistness, for 12 materials, such as powders, leather, and metal, using a visual analog scale (VAS), and confirmed relationships between tactile scores and multiple physical parameters measured with five instruments. These statistical confirmations of tactile scores and physical parameters have been applied in tactile design.
Recently, predictions of tactile perception based on physical parameters have been made using statistical and information methods. Hashim et al. [17] evaluated 26 automotive leathers using a hierarchical tactile vocabulary on a VAS scale, and measured compliance, vibration, temperature, and friction with four instruments to predict sensory and affective perceptions through multiple regression analysis. That model’s coefficient of determination (R2) ranged from 0.248 to 0.929, indicating high predictive accuracy for tactile descriptors such as temperature, hardness, and preference, but low accuracy for friction-related descriptors such as micro-roughness and macro-roughness. However, their approach was limited and lacked generalizability to other material groups, as it focused on specific materials, such as leather. Furthermore, the time and cost issues were not adequately addressed, as force, vibration, and temperature had to be measured using different instruments to predict multiple tactile descriptors, such as sensations and emotions.
Human fingers with active touch movements (e.g., pressing and stroking) are valuable tools for perceiving and recognizing physical parameters regardless of material type [18]. In recent years, studies have been conducted to estimate tactile perceptions using finger-shaped devices [19,20,21,22]. One such example is the Toccare system, which uses the BioTac [21], a finger-shaped device that simulates human tactile movements to touch materials [22]. This system allows the measurement of multiple physical parameters, such as force, vibration, and thermal stimulation, with a single device, without being limited by the type of material. Fishel et al. [23] used a finger-shaped device to measure force, vibration, and temperature across 117 different materials, with the results confirming the device’s high accuracy in evaluating roughness and friction. Systems that include this biomimetic multimodal tactile sensor have been used for evaluating the tactile sensations of liquid cosmetics on hair [24] and the surface roughness of plates with different treatments [25]. Furthermore, it has already been established that the BioTac can estimate sensory tactile perceptions [26]. However, even with a finger-shaped device, human affective tactile perceptions have not yet been accurately predicted. Affective tactile perceptions are complex and are composed of multiple physical properties and sensory tactile perceptions [13]. Therefore, we hypothesized that the Toccare system, which uses the multimodal tactile sensor to perform measurements simulating human touch movements, would also be capable of predicting affective tactile perceptions.
Given this background, in the present study, we focused on the sensory and affective tactile perceptions of various materials—primarily soft materials such as cosmetics, textiles, leather, wood, and rubber. Using a system that includes a biomimetic multimodal tactile sensor [22], we sought to obtain physical parameters through integrated measurements that simulate human touch movements. These physical parameters are the explanatory variables to be predicted, whereas the evaluation scores of our newly developed tactile vocabulary [11] are the objective variables. By examining 10 regression models to determine the model with the highest accuracy across various tactile descriptors, we aimed to predict each sensory and affective tactile perception more efficiently and affordably than conventional methods by using physical parameters. Our findings could be expected to be useful for more efficient tactile design and the creation of new tactile sensations in less time and at a lower cost compared with conventional methods.

2. Materials and Methods

2.1. Materials

In a previous study, 32 materials were evaluated (Table 1) [11]. That study primarily included soft materials that people frequently contact in daily life, such as cosmetics, fabrics, polymers, paper, rubber, wood, metal, tiles, clay, adhesive tape, and leather. All objects were prepared in 10 cm × 10 cm square shapes. Cosmetics, liquid lotions, serums, and creams were evaluated under two conditions. In the first method, the study subjects evaluated tactile perceptions when 0.2 g of a cosmetic product was applied directly onto a 10 cm × 10 cm square piece of artificial leather. In the second method, the evaluation occurred at 5 min after application.

2.2. Tactile Vocabulary

In this study, a tactile vocabulary comprising 31 descriptors was organized into two hierarchical layers: sensory and affective (Table 2) [11]. The sensory layer was composed of 19 descriptors describing the physical parameters of materials such as soft, smooth, and warm sensations, while the affective layer was composed of 12 descriptors expressing impressions and emotions evoked by the tactile perception, such as pleasant, like, and luxury.

2.3. Participants

The study participants were 64 Japanese native speakers (50% male; age range, 20–59 years) with an equal age distribution. All participants were informed about the content of the experiment and provided written consent to participate in the study and for their data to be used for research purposes. This study was reviewed and approved in advance by the Ethics Committee of POLA Chemical Industries, Inc. (Yokohama, Japan) (protocol code 2020-F-121, 21 October 2020).

2.4. Evaluation of Tactile Vocabulary Scores

In this study, the intensity scores for tactile vocabulary obtained in a previous study were used as the tactile perception scores [11]. The scores were defined as the median of the raw evaluation scores of the 64 subjects for each material. The participants touched one of each material and rated each tactile perception on a monitor using a six-point Likert-type scale, from 1, “not felt at all” to 6, “felt very strongly”. These evaluations were conducted under conditions where the participants could visually observe the materials to replicate the tactile evaluation humans experience in daily life. The subjects touched the surface of each material with the index finger of their nondominant hand. The subjects cleaned their fingers with wet wipes and tissues before touching different materials to ensure uniform finger conditions. The order of the material presentation and vocabulary was randomized. Each subject rated 10 materials each so that 20 people rated each material.

2.5. Evaluation of Physical Parameters

The physical parameters of the materials were evaluated using the Toccare system [22], which incorporates the BioTac biomimetic multimodal tactile sensor [21]. Figure 1 shows an overview of the system and device. The device was equipped with internal sensors, including a pressure sensor, electrodes, and a thermistor. The finger-shaped device is covered with an elastomer-based artificial skin filled with incompressible conductive fluid. Force is estimated based on voltage changes caused by the deformation of the artificial skin and the internal fluid when in contact with a material [27]. Micro-vibrations generated by contact with the material propagate through the internal fluid and reach a hydro-acoustic pressure sensor embedded in the sensor. The fingerprint pattern engraved on the surface of the artificial skin amplifies the vibration amplitude, enabling the detection of micro-vibrations [28]. The internal fluid is heated to human body temperature by a heater embedded in the electronic circuit. This ensures that, upon contact with a material, the sensor generates thermal conduction similar to human touch, allowing the estimation of the material’s thermal properties. From these measurement data, the original physical parameters listed in Table 3 were predicted using Bayesian exploration, resulting in 15 items similar to human tactile perception [24]. The 15 physical parameters were categorized as follows: friction (fST, fRS, and aTK), macro-roughness (mTX, mCO, and mRG), micro-roughness (μRO and μCO), thermal properties (tCO and tPR), and compliance (cCM, cDF, cDP, cRX, and cYD). Although the default lower and upper limits for each parameter were 0 and 100, respectively, scores of ≥100 were calculated in some cases [23,29]. Several defined movements—short- or long-distance sliding and pushing of the flat surface with the device—were performed in the evaluation of the physical parameters. The mean was selected as the physical parameter for each material when measurements were conducted five times.

2.6. Prediction of Tactile Perception Scores

Regression models for affective or sensory descriptors were examined using subject median scores for 15 physical parameters, 19 sensory tactile descriptors, and 12 affective tactile descriptors of 32 materials. A total of 10 estimation models were used, with eight regression models including AutoGluon [30], CatBoost [31], and Bagging Regressor [32]. The 10 regression models included four linear regression models: Linear Regression, Lasso, Ridge, and Elastic Net. As for nonlinear models, they consisted of SVR, Decision Tree, Random Forest, and ensemble learning models such as Gradient Boosting, CatBoost, and Bagging Regressor. Linear Regression is the most basic regression model, which represents the relationship between explanatory variables and the target variable using a linear function. Lasso improves model sparsity by adding an L1 regularization term to linear regression, which shrinks coefficients close to zero. Ridge prevents overfitting by adding an L2 regularization term to linear regression. Elastic Net combines Lasso and Ridge, incorporating both L1 and L2 regularization techniques. SVR (Support Vector Regression) applies Support Vector Machine (SVM) concepts to regression problems, performing regression by keeping data within a margin. The Decision Tree predicts outcomes by splitting data into multiple conditions, represented in a tree structure. Random Forest is an ensemble learning method that combines multiple decision trees to make predictions. Gradient Boosting constructs decision trees sequentially, learning iteratively to minimize prediction errors. CatBoost is a variant of gradient boosting, specialized for handling categorical data efficiently. The Bagging Regressor uses Bootstrap Aggregating to improve the robustness and accuracy of predictions. It works by creating multiple subsets of the original dataset through random sampling with replacement. Each subset is then used to train an independent model, typically of the same type, such as decision trees. The final prediction is obtained by averaging the predictions of all individual models. This approach reduces variance and prevents overfitting, which are common issues with individual models.
For sensory perception, the 15 physical parameters for the 32 materials were used as explanatory variables to create models for the 19 sensory scores. For affective scores, the same 15 physical parameters were used as explanatory variables to create models for 12 affective tactile scores. During model construction, the dataset was divided into training and test data at a ratio of 8:2. Hyperparameters were tuned to determine the best conditions for each target variable and maximize accuracy. This process was repeated five times with different data splits for cross-validation. Model accuracy was evaluated using R2. For each target variable, the R2 value of the model for the test data was evaluated, and the average R2 value across the five repetitions was used to represent the model accuracy. The criteria for model achievement were based on the accuracy reported in a previous study [17], with an average R2 value of ≥0.6. Regression models were developed using the Python 3.9 libraries autogluon.tabular, catboost, and sklearn.ensemble. Furthermore, the contribution of explanatory variables was calculated for each sensory and affective tactile perception prediction model.

3. Results

3.1. Prediction Models for Tactile Perception

The 15 physical parameters for each material are shown in Figure 2 and Table S1, the median of the participant sensory ratings for 19 descriptors in Table S2, and the median participant affective ratings for 12 descriptors in Table S3. Table 4 presents the test results of 10 regression models using the 15 physical parameters to predict each sensory and affective descriptor, consisting of four linear models, such as linear regression and ridge regression, and six nonlinear models, such as Support Vector Regression (SVR) and the Decision Tree. For each model, the number indicates the number of tactile descriptors for which R2 was ≥0.6 in the 19 sensory and 12 affective descriptors.
Among the linear models, Elastic Net had the highest number of descriptors with R2 values > 0.6: fourteen of nineteen for sensory descriptors and four of twelve for affective descriptors. By contrast, Ridge Regression had the lowest number, with only four of thirty-one descriptors higher than the threshold. For the nonlinear models, the Bagging Regressor had the highest number of descriptors with R2 values > 0.6: fourteen of nineteen for sensory descriptors and eight of twelve for affective descriptors. By contrast, SVR had the lowest number: three sensory descriptors and three affective descriptors. Overall, the Bagging Regressor was the top-performing model, covering the highest number of tactile descriptors with R2 values > 0.6 across all models for both sensory and affective descriptors.

3.2. Accuracy of Tactile Perception Prediction

The R2 and contribution of explanatory variables for the Bagging Regressor test data for the 19 sensory descriptors is shown in Table 5A. For the six descriptors representing moistness, the R2 values were 0.911 (slimy), 0.904 (soggy), 0.850 (sticky), 0.848 (wet), 0.823 (moist), and 0.748 (dry), indicating generally high estimation accuracy. For the two descriptors representing temperature, the R2 values were 0.901 (cold) and 0.337 (warm), showing high and low estimation accuracy, respectively. For the four descriptors representing elasticity and softness, the R2 values were 0.869 (elastic), 0.829 (soft), 0.796 (firm), and 0.767 (hard), showing generally high estimation accuracy. For the seven descriptors representing roughness and friction, the R2 values were 0.660 (bulky), 0.607 (slippery), 0.603 (rough), 0.592 (smooth), 0.487 (scratch), 0.484 (sleek), and 0.292 (bumpy). While three of the descriptors exceeded an R2 value of 0.6, four descriptors did not. The contribution of explanatory variables was identified for each tactile prediction model, listing the top three contributing physical properties out of the 15 parameters.
The results for the bagging regressor in predicting the 12 affective tactile descriptors are shown in Table 5B. The R2 values for the test data of each model were 0.829 (unpleasant), 0.705 (calm), and 0.627 (pleasant) for three descriptors related to comfort, discomfort, and arousal. For the two descriptors representing preference for a material, the R2 values were 0.676 (like) and 0.489 (dislike). For the seven descriptors representing impressions of the material, the R2 values were 0.774 (interest), 0.674 (luxury), 0.624 (fine), 0.622 (friendly), 0.516 (comfort), 0.485 (delicate), and 0.304 (slight warmth).
Scatterplots of the predicted and actual scores for the two tactile descriptors with the highest R2 values from the sensory and affective layers are shown in Figure 3 and Figure 4, respectively. Circles represent the training data, crosses represent the test data, and the dotted line indicates the one-to-one relationship between predicted and actual scores. Figure 3 shows the representative results for slimy and soggy as higher R2, and for bumpy and warm, which showed lower R2 for sensory descriptors. Figure 4 shows the representative results for unpleasant and interest as higher R2, and for slight warmth and delicate, which showed lower R2 for affective descriptors. For these tactile descriptors with higher R2, within the score range of 1–6, the predicted values increase as the actual values increase for both the training and test data, indicating a linear relationship.

4. Discussion

In this study, we evaluated 15 physical parameters of a diverse set of 32 materials, including cosmetics, textiles, leather, wood, metal, and rubber, using a system that included a biomimetic multimodal tactile sensor. Based on the obtained parameters, we succeeded in predicting accurate scores for 14 sensory descriptors and eight affective descriptors by the Bagging Regressor. Therefore, the present method appears to be useful for solving cost and time issues, as physical parameters of materials in different fields can be acquired by a single device and a wide range of sensory and affective perceptions can be predicted. The 32 materials included 26 soft materials selected from the 117 previously examined by Fishel et al. [23]. We added liquid materials such as cosmetics to broaden the range of commonly used materials. Previous studies have shown that a system with a biomimetic multimodal tactile sensor is suitable for the evaluation of a wide range of materials [24,25,26,33,34,35].
In this study, we examined 10 regression models using 15 physical parameters as explanatory variables to predict 19 sensory and 12 affective descriptors. We evaluated the average coefficient of determination for each tactile descriptor. We set the threshold for prediction as an average coefficient of determination of ≥0.6 based on the accuracy levels reported in a previous study [17]. Among the models, the largest number of tactile descriptors satisfied the threshold condition in the case of the Bagging Regressor. Previously, linear models were frequently used to predict tactile perceptions based on physical parameters [17]. Sagara et al. [36] recommended nonlinear models for predicting tactile perceptions, as human tactile characteristics are inherently nonlinear, and demonstrated that nonlinear models improve predictive accuracy compared with linear models. The bagging regressor is one of the nonlinear and ensemble learning methods. Ensemble learning includes techniques such as bagging, boosting, and stacking, all of which combine multiple models to achieve high accuracy and generalization [37]. Among these, the Bagging Regressor adopts the bagging approach, where multiple models are built, and their prediction results are averaged to make the final estimation. This model has been widely applied in fields such as real estate price prediction and economic growth forecasting [38,39,40,41,42,43]. The reason why the case using the bagging regressor achieved the highest accuracy in this study is likely because the approach of averaging multiple models was effective in addressing challenges such as the nonlinearity of tactile perception and noise caused by individual differences. By exploring such methods, we can improve the accuracy of tactile prediction and potentially show the mechanisms behind tactile perceptions.
Traditionally, sensory perception is known to be divided into five haptic dimensions [44,45,46,47,48]. As shown in Table 5A, the present model achieved high predictive accuracy for wetness and hardness and cold sensations. In addition, warmth was more difficult and roughness and bumpiness were more challenging to predict. The high predictive accuracy for wetness and hardness can be attributed to the larger variability in tactile evaluation scores among the materials. For cold sensations, the variability in evaluation scores was higher, particularly for liquid cosmetics. However, warmth showed smaller variability across materials, and the threshold between individual materials was insufficient; these trends resulted in lower predictive accuracy. For descriptors related to roughness and bumpiness, previous studies have shown that descriptors of roughness are correlated with the surface roughness parameter (Ra) [16]. Therefore, the physical measurements obtained in the present study may not fully capture the physical information related to the roughness perception. In the future, an increase in the diversity and number of materials could improve the prediction accuracy. Furthermore, the extraction of more relevant features related to roughness and unevenness from the vibration information obtained by the biomimetic multimodal tactile sensor will be important.
As shown in Table 5B, the predictive accuracy for affective descriptors was generally lower than that for sensory descriptors: a coefficient of determination of ≥0.6 was achieved for eight affective descriptors. Traditionally, attempts to estimate sensory tactile perceptions using other finger-shaped sensors or the BioTac have been made [21,22,24,26]. The ability to accurately predict affective tactile perceptions, as achieved in this study, holds significant importance for tactile perception prediction. Sufficient predictive accuracy was obtained for descriptors related to valence and arousal in Russell’s circumplex model of affection (“unpleasant”, “pleasant”, and “calm”) [49]. Furthermore, descriptors associated with personal preference (“like”) and those reflecting impressions of materials (“interest”, “luxury”, “fine”, and “friendly”) also showed high predictive accuracy. Affective tactile perception can be attributed to the complex interplay of multiple factors [13]. Furthermore, affective tactile perception is thought to be based on the hierarchical structure of physical parameters and sensory tactile perception [7]. The BioTac, being a multimodal sensor, simultaneously measures force, vibration, and temperature, accounting for the impact of each parameter. These characteristics of the tactile sensor may make it particularly effective for estimating complex tactile perceptions. However, the predictive accuracy was lower for descriptors such as “comfort”, “dislike”, “delicate”, and “slight warmth”. The low accuracy could be attributed to the uneven distribution of evaluation scores within the materials and the possibility that the physical measurements obtained in the present study may not fully capture the physical information related to above affective perceptions. Indeed, Figure 2 showed that parameters such as mCO, which represents macro-roughness, and tCO and tPR, which relate to temperature, have relatively small variance. As shown in Table 5, tactile prediction models with lower accuracy often included explanatory variables with small variance among their contributing physical parameters. For the affective descriptors that could not be predicted in this study, the features related to macro-roughness or thermal properties may have played an important role. Therefore, further investigation with a larger variety and quantity of materials is necessary to improve the prediction of affective descriptors. Additionally, affective perceptions are influenced by cultural backgrounds [50,51]. Therefore, further studies are needed to predict evaluation scores for participants with different cultural backgrounds or who do not speak Japanese as their native language. Furthermore, the variety and number of materials could be improved. An increase in the variation of materials could enhance the models that have lower predictive accuracy. The individual differences in evaluation scores were greater for affective tactile than for sensory descriptors. Sagara et al. [36] demonstrated that clustering participants based on their preferences and experiences improved predictive accuracy. In the present study, clustering participants has a potential for enabling the development of products tailored more closely to consumer preferences.

5. Conclusions

In this study, we examined a prediction model for tactile evaluation score by using a tactile vocabulary that enables tactile evaluation across various materials, guided by a suitability indicator, and by employing a biomimetic multimodal tactile sensor. This system mimicked human touch movements to measure physical parameters related to force, vibration, and temperature. The results confirmed that the model could predict sensory perceptions such as moistness, hardness, elasticity, and coldness, as well as affective perceptions such as pleasantness, unpleasantness, calmness, and preference. Based on our findings, more efficient tactile design and novel tactile experiences could be possible in the development of materials and products in various industrial sectors.

Supplementary Materials

The following supporting information can be downloaded at https://www.mdpi.com/article/10.3390/s25010147/s1. Table S1: Physical parameters of each material. Table S2: Sensory descriptor scores for each material. Table S3: Affective descriptor scores for each material.

Author Contributions

Conceptualization, T.I. and K.M.; Data curation, T.I.; Methodology, T.I. and K.M.; Investigation, T.I.; Writing—Original Draft Preparation, T.I.; Writing—Review and Editing, K.M. and Y.N.; Visualization, T.I.; Supervision, K.M. and Y.N. All authors have read and agreed to the published version of the manuscript.

Funding

This research was funded by POLA Chemical Industries, Inc.

Institutional Review Board Statement

This study was conducted according to the guidelines of the Declaration of Helsinki, and approved by the Ethics Committee of POLA Chemical Industries, Inc. (protocol code 2020-F-121, 21 October 2020).

Informed Consent Statement

Informed consent was obtained from all subjects involved in the study.

Data Availability Statement

The data that support the findings of this study are available on request from the corresponding author.

Conflicts of Interest

Authors Toshiki Ikejima and Koji Mizukoshi were employed by the company POLA Chemical Industries, Inc. The remaining author declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

References

  1. Citrin, A.V.; Stem, D.E.; Spangenberg, E.R.; Clark, M.J. Consumer need for tactile input. J. Bus. Res. 2003, 56, 915–922. [Google Scholar] [CrossRef]
  2. McCabe, D.B.; Nowlis, S.M. The effect of examining actual products or product descriptions on consumer preference. J. Consum. Psychol. 2003, 13, 431–439. [Google Scholar] [CrossRef]
  3. Peck, J.; Childers, T.L. To have and to hold: The influence of haptic information on product judgments. J. Mark. 2003, 67, 35–48. [Google Scholar] [CrossRef]
  4. Veelaert, L.; Du Bois, E.; Moons, I.; Karana, E. Experiential characterization of materials in product design: A literature review. Mater. Des. 2020, 190, 108543. [Google Scholar] [CrossRef]
  5. Piselli, A.; Baxter, W.; Simonato, M.; Del Curto, B.; Aurisicchio, M. Development and evaluation of a methodology to integrate technical and sensorial properties in materials selection. Mater. Des. 2018, 153, 259–272. [Google Scholar] [CrossRef]
  6. Zuo, H.; Jones, M.; Hope, T.; Jones, R. Sensory perception of material texture in consumer products. Des. J. 2016, 19, 405–427. [Google Scholar] [CrossRef]
  7. Drewing, K.; Weyel, C.; Celebi, H.; Kaya, D. Feeling and feelings: Affective and perceptual dimensions of touched materials and their connection. In Proceedings of the IEEE World Haptics Conference, Munich, Germany, 6–9 June 2017; pp. 25–30. [Google Scholar]
  8. Guest, S.; Dessirier, J.M.; Mehrabyan, A.; McGlone, F.; Essick, G.; Gescheider, G.; Fontana, A.; Xiong, R.; Ackerley, R.; Blot, K. The development and validation of sensory and emotional scales of touch perception. Atten. Percept. Psychophys. 2011, 73, 531–550. [Google Scholar] [CrossRef]
  9. Nagano, H.; Okamoto, S.; Yamada, Y. Modeling semantically multilayered affective and psychophysical responses toward tactile textures. IEEE Trans. Haptics 2018, 11, 568–578. [Google Scholar] [CrossRef] [PubMed]
  10. Osgood, C.E.; Sugi, G.J.; Tannenbaum, P.H. The Measurement of Meaning; University of Illinois Press: Urbana, IL, USA, 1957. [Google Scholar]
  11. Ikejima, T.; Mizukoshi, K.; Nonomura, Y. Development of a tactile vocabulary based on suitability and perceptual intensity. J. Sens. Stud. 2024, 39, e12956. [Google Scholar] [CrossRef]
  12. Lawless, H.T.; Heymann, H. Sensory evaluation of food. In Sensory Evaluation of Food—Principles and Practices; Food Science Text Series; Springer: New York, NY, USA, 2010; pp. 433–449. [Google Scholar]
  13. Chen, X.; Shao, F.; Barnes, C.; Childs, T.; Henson, B. Exploring relationships between touch perception and surface physical properties. Int. J. Des. 2009, 3, 67–76. [Google Scholar]
  14. Kim, W. A study on the subjective feeling affecting tactile satisfaction of leather in automobile: A structural equation modeling approach. Int. J. Ind. Ergon. 2021, 84, 103167. [Google Scholar] [CrossRef]
  15. Watanabe, S.; Horiuchi, T. Modeling perceptions using common impressions: Perceptual “authenticity”, “luxury”, and “quaintness” for leather. Text. Res. J. 2021, 91, 73–86. [Google Scholar] [CrossRef]
  16. Kikegawa, K.; Kuhara, R.; Kwon, J.; Sakamoto, M.; Tsuchiya, R.; Nagatani, N.; Nonomura, Y. Physical origin of a complicated tactile sensation: ‘shittori feel’. R. Soc. Open Sci. 2019, 6, 190039. [Google Scholar] [CrossRef]
  17. Mohamad Hashim, I.H.; Kumamoto, S.; Takemura, K.; Maeno, T.; Okuda, S.; Mori, Y. Tactile evaluation feedback system for multi-layered structure inspired by human tactile perception mechanism. Sensors 2017, 17, 2601. [Google Scholar] [CrossRef]
  18. Gibson, J.J. Observations on active touch. Psychol. Rev. 1962, 69, 477. [Google Scholar] [CrossRef]
  19. Sun, Y.; Liu, Y.; Pancheri, F.; Lueth, T.C. LARG: A Lightweight Robotic Gripper with 3-D Topology Optimized Adaptive Fingers. IEEE/ASME Trans. Mechatron. 2022, 27, 2026–2034. [Google Scholar] [CrossRef]
  20. Shan, X.; Birglen, L. Modeling and analysis of soft robotic fingers using the fin ray effect. Int. J. Robot. Res. 2020, 39, 1686–1705. [Google Scholar] [CrossRef]
  21. Fishel, J.A.; Loeb, G.E. Sensing tactile microvibrations with the BioTac—Comparison with human sensitivity. In Proceedings of the 2012 4th IEEE RAS & EMBS International Conference on Biomedical Robotics and Biomechatronics (BioRob), Rome, Italy, 24–27 June 2012; pp. 1122–1127. [Google Scholar]
  22. Technology, SynTouch. Available online: https://www.syntouchinc.com/wp-content/uploads/2017/07/WP_WorkingWithSynTouch_071017.pdf (accessed on 22 November 2024).
  23. Fishel, J.A.; Loeb, G.E. Bayesian exploration for intelligent identification of textures. Front. Neurorobot. 2012, 6, 4. [Google Scholar] [CrossRef]
  24. Liu, Z.; Graf, K.; Hub, J.; Kellermeier, M. Effects of cosmetic emulsions on the surface properties of Mongolian hair. ACS Omega 2022, 7, 10910–10920. [Google Scholar] [CrossRef]
  25. Skedung, L.; Harris, K.L.; Collier, E.S.; Rutland, M.W. The finishing touches: The role of friction and roughness in haptic perception of surface coatings. Exp. Brain Res. 2020, 238, 1511–1524. [Google Scholar] [CrossRef]
  26. Richardson, B.A.; Kuchenbecker, K.J. Learning to predict perceptual distributions of haptic adjectives. Front. Neurorobot. 2020, 13, 116. [Google Scholar] [CrossRef] [PubMed]
  27. Wettels, N.; Loeb, G.E. Haptic feature extraction from a biomimetic tactile sensor: Force, contact location and curvature. In Proceedings of the 2011 IEEE International Conference on Robotics and Biomimetics, Karon Beach, Thailand, 7–11 December 2011; pp. 2471–2478. [Google Scholar] [CrossRef]
  28. Fishel, J.A.; Santos, V.J.; Loeb, G.E. A robust micro-vibration sensor for biomimetic fingertips. In Proceedings of the 2008 2nd IEEE RAS & EMBS International Conference on Biomedical Robotics and Biomechatronics, Scottsdale, AZ, USA, 19–22 October 2008; pp. 659–663. [Google Scholar] [CrossRef]
  29. Xu, D.; Loeb, G.E.; Fishel, J.A. Tactile identification of objects using Bayesian exploration. In Proceedings of the 2013 IEEE international Conference on Robotics and Automation, Karlsruhe, Germany, 6–10 May 2013; pp. 3056–3061. [Google Scholar]
  30. Erickson, N. Autogluon-tabular: Robust and accurate automl for structured data. arXiv 2020, arXiv:2003.06505. [Google Scholar]
  31. Hancock, J.T.; Khoshgoftaar, T.M. CatBoost for big data: An interdisciplinary review. J. Big Data 2020, 7, 94. [Google Scholar] [CrossRef]
  32. Breiman, L. Bagging predictors. Mach. Learn. 1996, 24, 123–140. [Google Scholar] [CrossRef]
  33. Ruppel, P.; Jonetzko, Y.; Görner, M.; Hendrich, N.; Zhang, J. Simulation of the SynTouch BioTac Sensor. In Intelligent Autonomous Systems 15, Proceedings of the IAS 2018, Baden-Baden, Germany, 11–15 June 2018; Springer: Cham, Switzerland, 2019. [Google Scholar]
  34. Strand, M.; Dillmann, R.; Menegatti, E.; Ghidoni, S. (Eds.) Advances in Intelligent Systems and Computing; Springer: Cham, Switzerland, 2019; Volume 867, pp. 333–342. [Google Scholar] [CrossRef]
  35. Cao, L.; Russo, D.; Lapkin, A.A. Automated robotic platforms in design and development of formulations. AIChE J. 2021, 67, e17248. [Google Scholar] [CrossRef]
  36. Sagara, M.; Nobuyama, L.; Takemura, K. Nonlinear tactile estimation model based on perceptibility of mechanoreceptors improves quantitative tactile sensing. Sensors 2022, 22, 6697. [Google Scholar] [CrossRef] [PubMed]
  37. Dong, X.; Yu, Z.; Cao, W.; Shi, Y.; Ma, Q. A survey on ensemble learning. Front. Comput. Sci. 2020, 14, 241–258. [Google Scholar] [CrossRef]
  38. Breiman, L. Using Adaptive Bagging to Debias Regressions; Technical Report 547; Department of Statistics, UC Berkeley: Berkeley, CA, USA, 1999; p. 16. [Google Scholar]
  39. Graczyk, M.; Lasota, T.; Trawiński, B.; Trawiński, K. Comparison of bagging, boosting and stacking ensembles applied to real estate appraisal. In Intelligent Information and Database Systems, Proceedings of the Second International Conference, ACIIDS, Hue City, Vietnam, 24–26 March 2010; Proceedings, Part II 2; Springer: Berlin/Heidelberg, Germany, 2010; pp. 340–350. [Google Scholar]
  40. Altman, N.; Krzywinski, M. Ensemble methods: Bagging and random forests. Nat. Methods 2017, 14, 933–935. [Google Scholar] [CrossRef]
  41. Prasad, A.M.; Iverson, L.R.; Liaw, A. Newer classification and regression tree techniques: Bagging and random forests for ecological prediction. Ecosystems 2006, 9, 181–199. [Google Scholar] [CrossRef]
  42. Kadiyala, A.; Kumar, A. Applications of python to evaluate the performance of bagging methods. Environ. Prog. Sustain. Energy 2018, 37, 1555–1559. [Google Scholar] [CrossRef]
  43. Sharafati, A.; Asadollah, S.B.H.S.; Al-Ansari, N. Application of bagging ensemble model for predicting compressive strength of hollow concrete masonry prism. Ain Shams Eng. J. 2021, 12, 3521–3530. [Google Scholar] [CrossRef]
  44. Okamoto, S.; Nagano, H.; Yamada, Y. Psychophysical dimensions of tactile perception of textures. IEEE Trans. Haptics 2013, 6, 81–93. [Google Scholar] [CrossRef]
  45. Guest, S.; Mehrabyan, A.; Essick, G.; Phillips, N.; Hopkinson, A.; McGlone, F. Physics and tactile perception of fluid-covered surfaces. J. Texture Stud. 2011, 43, 77–93. [Google Scholar] [CrossRef]
  46. Summers, I.R.; Irwin, R.J.; Brady, A.C.; Grunwald, M. (Eds.) Haptic discrimination of paper. In Human Haptic Perception: Basics and Applications, 1st ed.; Springer: Berlin/Heidelberg, Germany, 2008; pp. 525–535. [Google Scholar]
  47. Yoshioka, T.; Bensmaıa, S.J.; Craig, J.C.; Hsiao, S.S. Texture perception through direct and indirect touch: An analysis of perceptual space for tactile textures in two modes of exploration. Somatosens. Mot. Res. 2007, 24, 53–70. [Google Scholar] [CrossRef]
  48. Tanaka, Y.; Tanaka, M.; Chonan, S. Development of a sensor system for measuring tactile sensation. In Proceedings of the 2006 IEEE Sensors, Daegu, Republic of Korea, 22–25 October 2006; pp. 554–557. [Google Scholar]
  49. Russell, J.A. A circumplex model of affect. J. Personal. Soc. Psychol. 1980, 39, 1161. [Google Scholar] [CrossRef]
  50. Mesquita, B.; Frijda, N.H. Cultural variations in emotions: A review. Psychol. Bull. 1992, 112, 179–204. [Google Scholar] [CrossRef]
  51. Kitayama, S.; Mesquita, B.; Karasawa, M. Cultural affordances and emotional experience: Socially engaging and disengaging emotions in Japan and the United States. J. Personal. Soc. Psychol. 2006, 91, 890–903. [Google Scholar] [CrossRef]
Figure 1. Overview of the Toccare system and biomimetic multimodal tactile sensor. (A) Experimental apparatus, (B) measuring mechanism, and (C) schematic of the BioTac finger-shaped tactile sensor.
Figure 1. Overview of the Toccare system and biomimetic multimodal tactile sensor. (A) Experimental apparatus, (B) measuring mechanism, and (C) schematic of the BioTac finger-shaped tactile sensor.
Sensors 25 00147 g001
Figure 2. Variation in physical parameters for materials. Gray zone is basic range measurement range. Each dot represents the score of a single material.
Figure 2. Variation in physical parameters for materials. Gray zone is basic range measurement range. Each dot represents the score of a single material.
Sensors 25 00147 g002
Figure 3. Prediction model for the representative sensory descriptors: (A) slimy, (B) soggy, (C) bumpy, and (D) warm. Blue circles indicate training data, and red crosses indicate test data.
Figure 3. Prediction model for the representative sensory descriptors: (A) slimy, (B) soggy, (C) bumpy, and (D) warm. Blue circles indicate training data, and red crosses indicate test data.
Sensors 25 00147 g003
Figure 4. Prediction model for the representative affective descriptors: (A) unpleasant, (B) interest, (C) slight warmth, and (D) delicate. Blue circles indicate training data, and red crosses indicate test data.
Figure 4. Prediction model for the representative affective descriptors: (A) unpleasant, (B) interest, (C) slight warmth, and (D) delicate. Blue circles indicate training data, and red crosses indicate test data.
Sensors 25 00147 g004
Table 1. List of materials used in this experiment.
Table 1. List of materials used in this experiment.
No.MaterialsNo.Materials
1Lotion *117Broad cloth
2Serum *118Cotton
3Cream *119Nylon
4Lotion (after application) *220Cashmere
5Serum (after application) *221Leather
6Cream (after application) *222Styrene foam
7Sticky tape23Western paper
8Slime24Japanese paper
9Clay25Wood plate
10Artificial skin26Stainless plate
11Fur27Polishing sponge
12Mesh (rough)28Cork
13Mesh (fine)29Convex rubber
14Sponge rubber30Tile
15Rubber31Acrylic plate
16Low rebound sponge32Artificial leather
*1: Material was applied to artificial leather (No. 32). *2: Material was applied to artificial leather (No. 32) and left for 5 min.
Table 2. List of descriptors in sensory and affective tactile vocabulary.
Table 2. List of descriptors in sensory and affective tactile vocabulary.
(A) Sensory Descriptors
No.Sensory Descriptors
1Bulky
2Bumpy
3Cold
4Dry
5Elastic
6Firm
7Hard
8Moist
9Rough
10Scratch
11Sleek
12Slimy
13Slippery
14Smooth
15Soft
16Soggy
17Sticky
18Warm
19Wet
(B) Affective Descriptors
No.Affective Descriptors
1Calm
2Comfort
3Delicate
4Dislike
5Fine
6Friendly
7Interest
8Like
9Luxury
10Pleasant
11Slight warmth
12Unpleasant
Table 3. Fifteen different physical parameters obtained by the Toccare system.
Table 3. Fifteen different physical parameters obtained by the Toccare system.
SensorsGenresItemsDetails
PressureFrictionfSTEffort required to initiate sliding on a surface
fRSEffort required to continue sliding on a surface
aTKEffort required to break contact with a surface
Macro-roughnessmTXIntensity of large features (>1 mm spacing)
mCOPerceived spacing of large features (>1 mm)
mRGPerceived uniformity of large features (>1 mm)
Micro-roughnessμROIntensity of small features (<1 mm spacing)
μCOPerceived spacing of small features (<1 mm)
ThermistorThermaltCOInitial rate that a surface draws heat from the fingertip
tPRExtent that a surface continues drawing heat from the fingertip
ElectrodesCompliancecCMDegree that a surface deforms under pressure
cDFDegree of surface wrap around the fingertip when being deformed
cDPSpeed that a surface returns to its original shape after being deformed
cRXDegree to which a surface stops pushing back after being deformed
cYDDegree to which a surface remains deformed after being pressed
Table 4. Comparison of prediction models for tactile descriptors.
Table 4. Comparison of prediction models for tactile descriptors.
ModelFeatureSensory
Descriptors
Affective
Descriptors
Total
Descriptors
Bagging RegressorNonlinear
(ensemble)
14/198/1222/31
Decision TreeNonlinear
(decision tree)
14/194/1218/31
Elastic NetLinear12/195/1217/31
Random ForestNonlinear
(decision tree)
11/196/1217/31
LassoLinear10/195/1215/31
Gradient BoostingNonlinear
(ensemble)
11/192/1213/31
CatBoostNonlinear
(ensemble)
8/193/1211/31
Linear RegressionLinear4/194/128/31
RidgeLinear4/194/128/31
SVRNonlinear3/193/126/31
The average coefficient of determination (R2) for each model was calculated for each objective variable. The numbers of sensory and affective descriptors represent how many objective variables had an average R2 of ≥0.6 in relation to the total number of objective variables. The total descriptors are the sum of the numbers for sensory and affective descriptors that had an average R2 of ≥0.6 across all descriptors.
Table 5. Accuracy of the tactile prediction model (Bagging Regressor).
Table 5. Accuracy of the tactile prediction model (Bagging Regressor).
(A) Sensory Descriptors
Sensory
Descriptors
Coefficient of DeterminationContribution of Explanatory Variables
Slimy0.911fRS, µCO, cDP
Soggy0.904aTK, cDP, tCO
Cold0.901tCO, cRX, cYD
Elastic0.869cRX, cDP, aTK
Sticky0.850aTK, cDP, fRS
Wet0.848fRS, cDP, tCO
Soft0.829cYD, cDP, cCM
Moist0.823cDP, cYD, fST
Firm0.796cDP, fST, cDF
Hard0.767fRS, tPR, cDF
Dry0.748cDP, fRS, µCO
Bulky0.660µCO, cDP, µRO
Slippery0.607µCO, cCM, mCO
Rough0.603cDP, µCO, tCO
Smooth0.592µCO, fST, µRO
Scratch0.487µRO, mCO, cRX
Sleek0.484cDF, mCO, µRO
Warm0.337tCO, mRG, mCO
Bumpy0.292mTX, fRS, fST
(B) Affective Descriptors
Affective
Descriptors
Coefficient of DeterminationContribution of Explanatory Variables
Unpleasant0.829aTK, µRO, cDF
Interest0.774fRS, cDP, aTK
Calm0.705µCO, cYD, mRG
Like0.676cCM, µRO, fRS
Luxury0.674µRO, fRS, cYD
Pleasant0.627µRO, aTK, cYD
Fine0.624µCO, µRO, aTK
Friendly0.622µCO, tCO, µRO
Comfort0.516µCO, tPR, µRO
Dislike0.489tCO, aTK, cRX
Delicate0.485tCO, aTK, tPR
Slight warmth0.304tCO, tPR, mRG
The average R2 of each tactile descriptor is shown. Bold font indicates an average R2 of ≥0.6. The contribution of explanatory variables is listed for each tactile prediction model, with the top three variables in order of their contribution.
Disclaimer/Publisher’s Note: The statements, opinions and data contained in all publications are solely those of the individual author(s) and contributor(s) and not of MDPI and/or the editor(s). MDPI and/or the editor(s) disclaim responsibility for any injury to people or property resulting from any ideas, methods, instructions or products referred to in the content.

Share and Cite

MDPI and ACS Style

Ikejima, T.; Mizukoshi, K.; Nonomura, Y. Predicting Sensory and Affective Tactile Perception from Physical Parameters Obtained by Using a Biomimetic Multimodal Tactile Sensor. Sensors 2025, 25, 147. https://doi.org/10.3390/s25010147

AMA Style

Ikejima T, Mizukoshi K, Nonomura Y. Predicting Sensory and Affective Tactile Perception from Physical Parameters Obtained by Using a Biomimetic Multimodal Tactile Sensor. Sensors. 2025; 25(1):147. https://doi.org/10.3390/s25010147

Chicago/Turabian Style

Ikejima, Toshiki, Koji Mizukoshi, and Yoshimune Nonomura. 2025. "Predicting Sensory and Affective Tactile Perception from Physical Parameters Obtained by Using a Biomimetic Multimodal Tactile Sensor" Sensors 25, no. 1: 147. https://doi.org/10.3390/s25010147

APA Style

Ikejima, T., Mizukoshi, K., & Nonomura, Y. (2025). Predicting Sensory and Affective Tactile Perception from Physical Parameters Obtained by Using a Biomimetic Multimodal Tactile Sensor. Sensors, 25(1), 147. https://doi.org/10.3390/s25010147

Note that from the first issue of 2016, this journal uses article numbers instead of page numbers. See further details here.

Article Metrics

Back to TopTop