Research on tactile perception by skin friction based on a multimodal method

Abstract

Background

Tactile perception is an essential function of skin. As this research involves many fields, such as skin friction, psychology, and neuroscience, the achievement tactile perception is scattered in various fields with different research methods. Therefore, it is necessary to study the whole tactile loop in a multimodal way, synchronizing all tactile information.

Materials and methods

To measure information from touch to haptics, we developed a specially designed measuring platform connecting to an electroencephalogram (EEG) recording system. Sandpapers with different roughness were used as samples. First, the surface properties were measured in tribological experiments. Second, psychophysical experiments were conducted to assess the volunteers’ cognition of samples’ roughness. Third, the mechanical parameters and EEG were measured at the same time during fingertip sliding on samples. Then, the data of all four tactile elements were processed and analyzed separately. The characteristic features were extracted from those data in the time‐frequency domain. Furthermore, the correlation coefficient was calculated in the pairwise comparison of each element to evaluate the feasibility of the multimodal method in the study of tactile perception.

Results

The 600‐mesh sandpaper has the largest Ra, Rz, Rsm, and particle size. The normal load, friction force, spectral centroid, and α‐ and β‐wave energy ratios of EEG at chosen electrodes have significant differences and correlations between 3000‐ and 600‐mesh sandpaper in general.

Conclusion

This multimodal method could be used in the study of tactile perception, which is a comprehensive way to observe the whole tactile loop from multiple perspectives.

1. INTRODUCTION

In the past decade, an increasing number of studies have shown great concern for the relationship between skin friction and tactile perception, ¹ because of its great significance in artificial intelligence. As is well known, the current artificial intelligence cannot understand the meaning of the external physical world, which is called the “physical grounding problem.” The Confucian perspective of “studying the nature of things” expresses that one must have physical contact to distinguish and identify things in order to obtain wisdom and intelligence. It has also been found that both tactile perception and proprioception are mediated by the protein PIEZO2 in recent research. ² It can be seen that the acquisition of tactile cognition is a necessary condition for the intelligent robot to have independent self‐awareness and realize intelligence. How is tactile cognition to be studied? The methods should not only be confined to the thought experiment of cognitive philosophy, but should also include positive philosophy.

Depending on the information transmitted to the brain, tactile sense can be divided into frictional haptics (sliding sense of touch) and haptics of muscle movement. Frictional haptics enables people to recognize objects without any other sensations. Frictional haptics originates from the skin's response to the mechanical stimulus generated by the object's properties. As shown in Figure 1, the skin stress–strain by friction is coded and transmitted via nerve impulse to form in the cerebral cortex. Thus, there should be a close correlation among every step of this tactile loop, including object properties, skin friction, brain physiology response, and cognition.

FIGURE 1.

Schematic diagram of the tactile loop

Many researchers have investigated the relationship between those steps. ³ , ⁴ , ⁵ , ⁶ , ⁷ , ⁸ , ⁹ , ¹⁰ , ¹¹ , ¹² Behmann ¹³ proposed that the higher the pressure on the surface of the fabric, the stronger the roughness. Ramachandran and Brang ¹⁴ found that samples with different textural characteristics (e.g., denim, wax, sandpaper, silk, etc.) evoked different emotions (e.g., depression, embarrassment, relief, and satisfaction). Although the psychophysical method is a convenient way to study the expression of touch, human feelings are very rich and subtle. The vocabulary to express feelings is similar in meaning and scarce. And there is often no perfect word to describe it, which reduces the effectiveness of psychophysical methods to some extent. Fagiani et al. ¹⁵ , ¹⁶ studied the frictional vibration process of skin with different surface roughness, conducted a spectral analysis of the vibration signals, and obtained information related to the physical properties of contact surfaces. Ding and Bhushan ⁷ applied an acceleration sensor to collect the vibrational signals generated by skin friction and discussed the difference in tactile textures between coated skin and uncoated skin. Dépeault et al. ¹⁷ studied the roughness of periodic and aperiodic equivalent samples with different longitudinal intervals using the scale method and found that periodicity had no obvious effect on the roughness. In our previous studies, we discussed the relationship between object properties and human subjective evaluation, ¹⁸ , ¹⁹ between frictional vibration and surface properties, ²⁰ between skin structures and friction, ²¹ , ²² , ²³ , ²⁴ and between friction and electroencephalogram (EEG). ¹⁸ , ²⁵ But as of now, few studies have been reported on the entire tactile loop; that's probably because this study requires multiple methods with the combination of different research fields.

Based on other studies, ¹⁷ , ²⁶ , ²⁷ a multimodal platform was developed to measure surface topography, skin friction, EEG, and cognition in this paper. Tactile roughness was observed in this paper since it is an accepted tactile dimension. ²⁸ , ²⁹ , ³⁰ , ³¹ EEG, commonly used to observe the brain's physiological response to tactile stimulus, ⁴ , ¹⁸ , ²⁵ , ³² , ³³ was recorded synchronously with other information from surface topography, skin friction, and cognition. The total of 16 parameters were extracted to characterize these multimodal signals, and Pearson correlation analysis was conducted.

2. EXPERIMENTAL DETAILS

The experiment passed the medical ethics review and was conducted following the moral standards stipulated in the Declaration of Helsinki. All the experimental volunteers were informed of the experimental details and cautions. They all signed the informed consent and had the right to withdraw from the experiment at any time during the experiment. Experiments were performed by 10 males and 10 females, all healthy right‐handed volunteer graduate students, aged between 19 and 23.

Sandpapers with different roughness were selected as the experimental samples: 3000 mesh, 1000 mesh, 800 mesh, and 600 mesh. Mesh is generally defined as the number of holes in the screen within a 1‐inch line segment; the more the mesh value, the more exquisite the texture. The ambient temperature was 24 ± 1°C and the relative humidity was 50% ± 5%.

Four experiments were carried out in this paper: (1) surface topographic experiments to study the surface properties of sandpaper with different roughness, (2) psychological physics experiments to get a subjective evaluation on samples according to the dimension of rough/smooth, (3) tribology experiments to get frictional and vibrational data, and (4) nerve electrophysiology experiments to observe brain activity by EEG data. Among them, the first two experiments were carried out separately, while the second two experiments were carried out simultaneously on the same experimental platform.

2.1. Surface topographic test

The sandpaper samples were cut into square panels the size of 30 mm × 30 mm.

According to the microscopic contact theory model, the surface is composed of asperities, which can be described by some characteristic parameters of height and distance. A three‐dimensional optical morphometer (Contour GT‐InMotion) was used to detect the parameters of Ra, Rz, Rsm, and particle size for four mesh sandpaper samples. Each parameter of each sample was repeatedly measured three times in three directions of the 30 mm sampling length, and the average value was taken.

Ra represents the average height of the sample, which is the arithmetic mean deviation of absolute distance between the point on the contour line and the baseline, as described in Equation (1), where n is the sampling length.

The surface roughness height characteristic can also be described as the maximum height, Rz, which is the average distance between the adjacent peaks and valleys, as expressed in Equation (2).

Rsm represents the spacing characteristics. It is the average value of the combined spacing between the unimodal peak and its adjacent valley, which is approximate to Equation (3):

$$Ra=\frac{1}{n}\sum _{i=1}^n|y_i|$$ (1)

$$Rz=\frac{1}{n}\left(\sum _{i=1}^n{p}_i-\sum _{i=1}^n{v}_i\right)$$ (2)

$$Rsm=\frac{1}{n}\sum _{i=1}^{n}X{s}_i$$ (3)

2.2. Psychological physics experiment

The psychophysical experimental samples were tailored to the size of 100 mm × 100 mm.

The experimental details have been described by Chen et al. ¹⁹ In the tests, volunteer students with their eyes covered touched and slid their right thumb on samples. They gave their judgments on the texture dimensions of rough/smooth. The dimension grade ranked from 1 to 10, in which a higher grade represented a smoother surface. The final scores of each dimension were shown as the average of the scores given by all 20 volunteers for one sample in the formal experiments.

2.3. Tribology experiment and nerve electrophysiology experiment

Because EEG data is sensitive to stimuli, a tactile stimulus generator was developed, which could also synchronously measure multimode signals, including normal load, friction force, vibration, and EEG (Figure 2).

FIGURE 2.

Tactile perception multimodal information measuring device

The signal acquisition module consists of 15 pieces of soft pressure sensors, an accelerometer, and a torque sensor, respectively collecting active pressure, friction, and vibration signals. The communication transmits these three signals and uses the shaft encoder on the roller to perform position marking. When each sample is transposed to the specified starting point, a trigger signal is sent to the EEG device (EGI, 64‐channel, USA) via a synchronizer (DB9 Pinout). The EEG signals of each sample were marked. A sample substrate was made of silicone oil paper covering the pressure sensor all around the roller without interval. The sandpaper samples were cut into pieces measuring 690 mm × 50 mm, which were stuck onto the substrate. The accelerometer, which measured the vibration signal between the finger and the sample, was placed on the cover of the right index fingernail, as shown in Figure 3A,B.

FIGURE 3.

(A) Experiment scene diagram for tactile perception of multimodal measurement, (B) finger touching sandpaper samples, and (C) electrode distribution of EGI 64 electroencephalogram cap

In order to collect EEG data, the EEG cap was directly worn by the volunteers. The volunteers were asked to rest and wash their hair before the experiment. The EEG cap electrodes were arranged in the pattern of 10‐10 international system electrodes, as shown in Figure 3C. Then, electrolyte was injected into the electrode sponge with a dropper to keep the electrode impedance under 20 kΩ. The electrolyte solution was prepared in advance according to the following formula: 1.5 L pure water, two tablespoons of potassium chloride, and two teaspoons of shampoo (Johnson & Johnson baby shampoo).

After the finger touched the sample, the roller began to spin for one cycle at a linear velocity of 10 mm/s. This was repeated two times with a 1‐min interval. Then the sample was changed for the next experiment. The volunteers rested for 5 min for every sample.

3. DATA PROCESSING AND ANALYSIS

3.1. Frictional signal processing

The force sensors were scaled before the experiments by using a pressure meter. A fit curve was drawn for each force sensor. Also, the no‐load signal of force sensors was recorded when the roller was spinning. Then, the normal load could be determined with the fit curve and no‐load pressure. The normal load was averaged after preprocessing for various mesh sandpaper samples. The friction force was obtained from a torque sensor using the same method as with the force sensors. The typical normal load and friction force are shown in Figure 4.

FIGURE 4.

(A) Normal load and (B) friction force of different mesh sandpaper samples after pretreatment

3.2. Vibrational signal processing

The vibrational signal was obtained from an accelerometer. It was denoised by the ensemble empirical mode decomposition method. The original and denoised acceleration signals are shown in Figure 5A,B.

FIGURE 5.

(A) Original acceleration signal and (B) denoised acceleration signal

Then spectral centroid and average power were extracted from the denoised signals. ³⁴

3.3. EEG processing

EEG was preprocessed, including reference electrodes setting (mastoid area), band‐pass filtering (bandwidth was 1∼30 Hz), baseline removing (averaging 200 ms potentials before stimulus), bad electrode removing, electrooculogram, and myoelectricity removing. The typical preprocessed EEG induced by touching different mesh sandpapers is shown in Figure 6.

FIGURE 6.

Typical preprocessed electroencephalogram induced by touching various samples

Next, EEG rhythms were separated from each other by wavelet packet transform, as shown in Equation (4) where $h$、$g$ is the filter coefficient, $d$ is the wavelet packet decomposition coefficient, $p$、$t$ is the number of decomposition levels, and $j$、$k$ is the node number of the wavelet packet:

$$\left.\begin{array}{c}{\displaystyle d_t^{j,2k}=\sum _p{h}_{p-2t}{d}_p^{j+1,2k}}\hfill \\ {\displaystyle d_t^{j,2k+1}=\sum _p{g}_{p-2t}{d}_p^{j+1}}\hfill \end{array}\right\}$$ (4)

Then, α wave (8–14 Hz) and β wave (14–30 Hz) were reconstructed by using the selected wavelet decomposition level with Equation (5):

$$d_t^{j+1,k}=\sum _p\left(h_{t-2p}{d}_p^{j,2k}+g_{t-2p}{d}_p^{j,2k+1}\right)$$ (5)

Energy was extracted as the characteristics from the reconstructed α and β waves by using Equation (6), where $i$ is the index in the time domain of the reconstructed signal, $i=1:n$, and $j$ is the number of wavelet levels. The energy ratio of each rhythm was composed of the ratio of the rhythm energy to the total energy:

$$E_j=\sum _{i=1}^n{y}_j^2$$ (6)

Energy ratios were widely used to indicate the EEG rhythm characteristics under the stimulus. Horiba et al. ³⁵ proposed that the α wave can effectively evaluate tactile sensation. Nam et al. ³² showed that α and β waves in the contralateral brain hemisphere were inhibited during one‐handed use. Hammeke et al. ³⁶ found that the somatosensory cortex of the parietal lobe is sensitive to tactile stimulus and has a greater response to comfortable samples than to uncomfortable ones.

3.4. The statistical method and correlation analysis method of multimodal signals

The statistical analysis of this experiment was carried out on SPSS19.0 (IBM, USA). A paired t‐test was performed to test the significant difference between the four samples with regard to the measured surface topography, skin friction, brain activity, and tactile roughness. After completing the paired t‐test, we viewed the p‐value of the paired sample in the results. If p < 0.05, it meant that the results were significantly different.

To find the relationship from touch to haptics, Pearson correlation coefficient of those multimodal signals was analyzed because it reflects the linear correlation degree of two variables. ³⁷ The calculation formula is shown in Equation (7), where (x, y) represents two variable sets, and N represents the number of samples:

$$Pearson(x,y)=\frac{\sum xy-\frac{\sum x\sum y}{N}}{\sqrt{\left(\sum x^2-\frac{{\left(\sum x\right)}^2}{N}\right)\left(\sum y^2-\frac{{\left(\sum y\right)}^2}{N}\right)}}$$ (7)

Person's correlation coefficient returns a value between −1 and 1. If the correlation coefficient is 1, it indicates a strong positive relationship between the variables. If the correlation coefficient is 0, it indicates no relationship.

4. RESULTS AND DISCUSSION

4.1. Surface topography

The variation trend of surface properties of each object with sandpaper mesh is shown in Table 1. The 600‐mesh sandpaper has the largest Ra, Rz, Rsm, and particle size. The sandpaper of 3000 mesh has the minimum values.

TABLE 1.

Sandpaper surface roughness parameter comparison

Samples	3000 mesh	1000 mesh	800 mesh	600 mesh
Ra (μm)	3.2 ± 0.3	5.8 ±0.3	6.4 ± 0.3	8.1 ± 0.4
Rz (μm)	18.3 ± 1.7	33.4 ± 1.4	36.7 ± 2.1	45.0 ± 1.6
Rsm (μm)	15.6 ± 1.4	25.4 ± 1.7	31.7 ± 1.5	42.3 ± 2.2
Particle size (μm)	7.1 ± 0.6	13.4 ± 1.0	18.2 ± 1.2	23.8 ± 1.3

4.2. Results of psychological physics experiments and tribology experiments

The results of the psychological physics experiments are shown in Figure 7A. Compared to Figure 6, the tactile roughness scores showed the same increasing trend as the surface properties of sandpaper samples. Also, there is a slow increase from 1000 mesh to 800 mesh just as Ra and Rz. This could be explained by the results of the paired t‐test, which showed that there are significant differences in the tactile roughness scores of volunteers’ subjective evaluations for sandpaper samples, except that for 1000 and 800 mesh. That means the volunteers have similar tactile roughness on these two samples. In other words, it is not easy to distinguish between the two samples in this experiment leading to similar scores.

FIGURE 7.

(A) Tactile roughness score in psychological physics experiment, (B) normal load of skin friction, (C) friction force of skin friction, (D) spectral centroid of the vibrational signal, and (E) averaged power of the vibrational signal

Figure 7B shows that a larger normal load was applied to sandpaper samples of 1000 and 800 mesh than to that of 3000 and 600 mesh. The possible reason is their similar tactile feelings for 1000 and 800 mesh, the same as the above analysis. For 600 mesh, the discomfort in tactile feeling could also cause a small normal load due to its rough surface.

The friction force increased with increasing tactile roughness, as shown in Figure 7C. But it was not in line with this trend for a sample of 600 mesh because of the reduction of the real contact area due to a small normal load. The results of paired t‐tests showed that friction played an important role in characterizing surface properties.

Spectral centroid and the averaged power of vibrational signals were all affected by surface topography; as shown in Figure 7D,E, the spectral centroid decreased with increasing tactile roughness and Rsm. This is because the spectral centroid represents the vibrational frequency of fingertip skin and is largely related to surface spacing characteristics. A larger value of Rsm indicates fewer unimodal peaks leading to lower vibration frequency. The averaged power is largely related to surface height characteristics since it is proportional to the vibrational amplitude. It increased with increasing tactile roughness, Ra, and Rz. The paired t‐test shows that the spectral centroid is more suitable than the average power to measure the vibration signal when the finger touches sandpaper with different roughness since the results of averaged power have no significant differences in various samples.

4.3. Results of nerve electrophysiology experiment

In this part of the experiment, three volunteers dropped out of the experiment for personal reasons. Five volunteers complained of their poor status in EEG experiments. At last, the data of seven males and five females were recorded and analyzed.

The paired t‐test was carried out on the EEG data of all electrodes of different volunteers for various samples. The typical results of one volunteer and one pair of samples are shown in Figure 8.

FIGURE 8.

Typical results of paired t‐test of electroencephalogram rhythm energy ratio on all electrodes of some volunteer who touched sandpaper of 3000 mesh and 600 mesh

There were significant differences in the α‐wave energy ratio in nearly the whole brain, while for the β‐wave energy ratio it was mostly in the occipital lobe. The activated occipital lobe was assumed to be engaged in visual activity. Studies ³⁸ , ³⁹ , ⁴⁰ have also reported that the activated visual cortex area (especially the lateral occipital complex) is related to multichannel perception during tactile recognition. Therefore, it is speculated that visual perception is involved in this experiment, although the volunteers were wearing blindfolds during the whole experimental process. The activated occipital electrode may also indicate that the brain has already completed the perception of sandpaper roughness due to the multichannel perception effect of vision combined with touch.

According to Figure 8, seven electrodes on sagittal (AFz, FCz, Pz, and Oz) and coronal line (C3, Cz, and C4) were selected for further analysis on the samples of 3000 and 600 mesh to observe α‐ and β‐wave energy ratios along with changing tactile roughness in the time domain. As shown in Figure 9, the α‐ and β‐wave energy ratios of 3000‐mesh sandpaper are higher than that of 600‐mesh sandpaper. The intersample differences of the α‐wave energy ratio are larger than those of β‐wave energy ratios. That means the α‐wave energy ratio could be the parameter to characterize the brain activations related to tactile roughness.

FIGURE 9.

α‐ and β‐wave energy ratios for sandpaper samples of 3000 mesh and 600 mesh at electrode (A) FCz, (B) Cz, and (C) POz in the time domain

4.4. Results of correlations of multimode signals

In this experiment, Pearson bivariate correlation analysis was carried out among the multimode signals, including surface topography (Ra, Rsm, Rz, and particle size), skin friction (normal load, friction force, vibrational spectral centroid, and vibrational averaged power), brain activity (α‐ and β‐wave energy ratios of EEG at chosen electrodes were recorded briefly as FCz_α, FCz_β, Cz_α, Cz_β, POz_α, and POz_β), and tactile roughness (subjective evaluation scores). The results are shown in Tables 2, 3, and 4. Star superscript indicates the results of paired t‐test were significantly different. It can be seen from the above results that there is a connection between those multimodal signals since one or more modal eigenvalues were found to be correlated. That means it is feasible to study the tactile perception of friction by the multimode method.

TABLE 2.

Correlation analysis of surface topography towards tactile roughness and EEG features

(a) Correlation analysis of surface topography and tactile roughness
	Surface topography
Tactile roughness	Mesh	Ra	Rsm	Rz	Particle size
Subjective evaluation score	−0.957**	0.998**	0.972**	0.980**	0.976**

(b) Correlation analysis of surface topography and electroencephalogram (EEG) features
	Surface topography
EEG	Mesh	Ra	Rsm	Rz	Particle size
FCz_α	0.482*	−0.431*	−0.335*	−0.437*	−0.317
FCz_β	0.368*	−0.347*	−0.288	−0.35*	−0.275
Cz_α	0.525*	−0.446*	−0.340*	−0.455*	−0.327
Cz_β	0.426*	−0.414*	−0.35	−0.416*	−0.334
POz_α	0.436*	−0.380*	−0.293*	−0.386*	−0.28
POz_β	0.565*	−0.459*	−0.331	−0.471*	−0.317

(c) Correlation analysis of surface topography and skin friction
	Surface topography
Skin friction	Mesh	Ra	Rsm	Rz	Particle size
Averaged power	−0.125	−0.353	−0.201	−0.269	0.305
Spectral centroid	0.227	0.405	0.192	0.390	0.344
Normal load	0.217	0.412	0.193	0.379	0.352
Friction force	0.251	0.365*	0.155	0.428*	0.348

TABLE 3.

Correlation analysis of skin friction towards tactile roughness and EEG features

(a) Correlation analysis of skin friction and tactile roughness
	Skin friction
Tactile roughness	Averaged power	Spectral centroid	Normal load	Friction force
Subjective evaluation score	0.211	0.392	−0.375	−0.306*

(b) Correlation analysis of skin friction and electroencephalogram (EEG) features
	Skin friction
EEG	Averaged power	Spectral centroid	Normal load	Friction force
FCz_α	−0.056	−0.174	−0.126	−0.216*
FCz_β	0.146	−0.508	−0.231	−0.29*
Cz_α	−0.002	−0.22	−0.183	−0.194
Cz_β	0.16	−0.437	−0.299	−0.323*
POz_α	0.26	−0.238	−0.226	−0.256*
POz_β	0.256	−0.478	−0.226	−0.238*

TABLE 4.

Correlation analysis of electroencephalogram (EEG) features and tactile roughness

	EEG
Tactile roughness	FCz_α	FCz_β	Cz_α	Cz_β	POz_α	POz_β
Subjective evaluation score	−0.516*	−0.198	−0.430*	−0.345*	−0.320*	−0.139

Specifically, Table 2 shows the correlation analysis of surface topography with tactile roughness, EEG features, and skin friction. The tactile roughness is strongly correlated with surface topography, which proves that human tactile cognition is the perfect system to classify texture. The EEG features were correlated with most features of surface topography, except particle size, while tactile roughness had a strong correlation with particle size. The possible reason is that the volunteers’ task in the nerve electrophysiology experiment was to sense surface roughness. The simple task is suitable for multimodal experiments but it might lead to information loss. Besides, the α‐ and β‐wave energy ratios at electrodes FCz, Cz, and POz are moderately weak in relation to the mesh, Ra, and Rz. The α‐wave energy ratio is moderately weak in relation to Rsm. The α wave is related to feelings of comfort. ⁴¹ Sandpaper with different physical characteristics may have brought different levels of comfort to the subjects and therefore had some relevance. The β waves are attention‐related brain waves ⁴² that require a high degree of concentration when distinguishing sandpaper from different physical characteristics and therefore have a certain correlation. These results indicate that the multimodal experimental mode and EEG features should be discussed in more detail for further studies.

Table 2(c) and Table 3 show that friction force has correlation with surface topography, tactile roughness, and EEG features. However, there appeared to be poor correlation for other skin friction features from the vibration signal. According to successful cases, ¹⁵ , ⁴³ , ⁴⁴ the accelerometer fastened to the nail causes distortion of the vibration signals filtered by fingers’ connective tissue and muscle tissue. In our previous studies, ¹⁸ , ¹⁹ , ²⁰ we have also discussed the relationship of vibration signals, topography, and EEG features. Those results showed good relationship since the signals were recorded using a biomimetic finger, because of which the vibration is exactly from the interface. Therefore, the position of the accelerometer might be important and reconsidered for signal acquisition in multimodal experiments.

Table 4 shows that tactile roughness was negatively correlated with EEG features on several electrodes, which means those EEG features could not fully represent cognition. In addition, not only the more appropriate features, but also the high‐density EEG acquisition equipment are helpful for multimodal studies.

5. CONCLUSIONS

In this paper, a specially designed platform was developed to measure multimodal information on the human tactile loop. The characteristics of multimodal signals and the correlation between them were studied to discuss the feasibility of multimodal haptic measurement methods. Four experiments were done to get the multimodal signal at the same time: surface topography experiment, psychological physics experiments, tribology experiments, and nerve electrophysiology experiments. Data on surface topography, skin friction, brain activity, and tactile roughness were recorded and analyzed. Sixteen parameters were chosen to represent these multimodal signals, including surface topography (mesh, Ra, Rsm, Rz, and particle size), skin friction (normal load, friction force, vibrational spectral centroid, and vibrational averaged power), brain activity (α‐ and β‐wave energy ratios of EEG at chosen electrodes), and tactile roughness (subjective evaluation scores). The paired t‐test was applied in significance testing across multiple comparisons of different meshed sandpapers, and Pearson correlation coefficient of these 16 parameters was calculated. The results suggested that the normal load, friction force, spectral centroid, and α‐ and β‐wave energy ratios of EEG at the chosen electrodes had significant differences and correlations between 3000‐ and 600‐mesh sandpaper in general, which proved that the multimodal measurement method of tactile perception has certain feasibility.

The multimodal experiment is a time‐consuming process. It is not suitable for some volunteers, which might lead to a shortage of data. Also, this developed platform could be an interface for functional Near Infrared Spectroscopy and functional Magnetic Resonance Imaging; more information about brain activities are expected in further studies.

Chen S, Qiao X, Yang J, Ru W, Tang W, Zhang S. Research on tactile perception by skin friction based on a multimodal method. Skin Res Technol. 2022;28:280–290. 10.1111/srt.13127